【都灵理工】 工厂设计01-FIP_Principles

半导体厂房项目总平面设计思路发布时间：2023-01-13T07:35:49.849Z 来源：《建筑实践》2022年第18期作者：许林林[导读] 近些年来，随着我国国民经济的发展以及全球竞争力的提高，对各行各业的厂房建设都进行了一定的发展与改进许林林信息产业电子第十一设计研究院科技工程股份有限公司华东分院江苏省无锡市 214000摘要：近些年来，随着我国国民经济的发展以及全球竞争力的提高，对各行各业的厂房建设都进行了一定的发展与改进。

随着工业化的改革，在开展建筑工作之前进行平面设计应当详读规划，进行全局规划。

本文介绍半导体厂房项目总平面设计思路，针对厂房项目总平面方案设计做出系统性的分析。

关键词：厂房项目总平面设计思路工业企业总平面设计是项目建设当中重要的组成部分，没有完整的平面设计思路就会使整体的项目总体分散，会导致盲目的建设，既影响了建设的效果，又影响到总平面设计组织，破坏了建筑群体的稳定性。

因此，在厂房项目建设之前，应当确定总平面设计方案，合理的选取用地范围，做好协调工作。

一、总平面设计工业企业总平面设计就是指根据对厂房要求组成建筑群，并综合考虑建筑物以及各项设施之间的关系，科学看待厂房建筑存在的问题，充分的考虑并注重环保节约状况，使建筑群与周边环境组成内容成为统一的集体，将周边环境相协调，进行系统化的设计。

项目总平面设计通常需要考虑建筑物、构筑物、绿化与各种管线之间的关系，并且详细的考察施工地形、地质，选取适当的气候条件进行相应的布置，并且采用图纸对施工场地以及构筑物进行标注。

工厂项目总平面设计是一项技术性的优化设计，厂房项目追求的效果是多目标的综合效果，不仅是为了追求经济效益，还是将生态与环境、社会经济统一的整体综合效益。

因此，在任何工厂的项目建设设计方面，应当以工厂为主体，创设与工厂相适应的物质条件的综合体。

企业在项目建设方面，要善于研究各种气候条件或特殊条件，为工业化生产创造最佳的整体效益。

CHAPTER3ATOMIC COLLISIONS3.1BASIC CONCEPTSWhen two particles collide,various phenomena may occur.As examples,one or both particles may change their momentum or their energy,neutral particles can become ionized,and ionized particles can become neutral.We introduce the funda-mentals of collisions between electrons,positive ions,and gas atoms in this chapter, concentrating on simple classical estimates of the important processes in noble gas discharges such as argon.For electrons colliding with atoms,the main processes are elastic scattering in which primarily the electron momentum is changed,and inelas-tic processes such as excitation and ionization.For ions colliding with atoms,the main processes are elastic scattering in which momentum and energy are exchanged, and resonant charge transfer.Other important processes occur in molecular gases. These include dissociation,dissociative recombination,processes involving negative ions,such as attachment,detachment,and positive–negative ion charge transfer,and processes involving excitation of molecular vibrations and rotations. We defer consideration of collisions in molecular gases to Chapter8.Elastic and Inelastic CollisionsCollisions conserve momentum and energy:the total momentum and energy of the colliding particles after collision are equal to that before collision.Electrons and fully stripped ions possess only kinetic energy.Atoms and partially stripped ions have internal energy level structures and can be excited,de-excited,or ionized, Principles of Plasma Discharges and Materials Processing,by M.A.Lieberman and A.J.Lichtenberg. ISBN0-471-72001-1Copyright#2005John Wiley&Sons,Inc.43corresponding to changes in potential energy.It is the total energy,which is the sum of the kinetic and potential energy,that is conserved in a collision.If the internal energies of the collision partners do not change,then the sum of kinetic energies is conserved and the collision is said to be elastic.Although the total kinetic energy is conserved,kinetic energy is generally exchanged between particles.If the sum of kinetic energies is not conserved,then the collision is inelas-tic.Most inelastic collisions involve excitation or ionization,such that the sumof kinetic energies after collision is less than that before collision.However,super-elastic collisions can occur in which an excited atom can be de-excited by acollision,increasing the sum of kinetic energies.Collision ParametersThe fundamental quantity that characterizes a collision is its cross section s(v R), where v R is the relative velocity between the particles before collision.To define this,we considerfirst the simplest situation shown in Figure3.1,in which aflux G¼n v of particles having mass m,density n,andfixed velocity v is incident on a half-space x.0of stationary,infinitely massive“target”particles having density n g.In this case,v R¼v.Let d n be the number of incident particles per unit volume at x that undergo an“interaction”with the target particles within a differential distanced x,removing them from the incident beam.Clearly,d n is proportional to n,n g,and d x for infrequent collisions within d x.Hence we can writed n¼Às nn g d x(3:1:1)where the constant of proportionality s that has been introduced has units of area and is called the cross section for the interaction.The minus sign denotes removal from the beam.To define a cross section,the“interaction”must be specified,for example,ionization of the target particle,excitation of the incident particle to a given energy state,or scattering of the incident particle by an angle exceeding p=2.Multiplying(3.1.1)by v,wefind a similar equation for theflux:d G¼Às G n g d x(3:1:2) FIGURE3.1.Aflux of incident particles collides with a population of target particles in the half-space x.0.44ATOMIC COLLISIONSFor a simple interpretation of s,let the incident and target particles be hard elastic spheres of radii a1and a2,and let the“interaction”be a collision between the spheres.In a distance d x there are n g d x targets within a unit area perpendicular to x.Draw a circle of radius a12¼a1þa2in the x¼const plane about each target.A collision occurs if the centers of the incident and target particles fall within this radius.Hence the fraction of the unit area for which a collision occurs is n g d x p a212.The fraction of incident particles that collide within d x is thend G G ¼d nn¼Àn g s d x(3:1:3)wheres¼p a212(3:1:4)is the hard sphere cross section.In this particular case,s is independent of v.Equation(3.1.2)is readily integrated to give the collidedfluxG(x)¼G0(1ÀeÀx=l)(3:1:5) with the uncollidedflux G0eÀx=l.The quantityl¼1n g s(3:1:6)is the mean free path or the decay of the beam,that is,the distance over which the uncollidedflux decreases to1=e of its initial value G0at x¼0.If the velocity of the beam is v,then the mean time between interactions ist¼lv(3:1:7)Its inverse is the interaction or collision frequencyn;tÀ1¼n g s v(3:1:8)and is the number of interactions per second that an incident particle has with the target particle population.We can also define the collision frequency per unit density,which is called the rate constantK¼s v(3:1:9)3.1BASIC CONCEPTS45and,trivially,from (3.1.8)and (3.1.9)n ¼Kn g(3:1:10)Differential Scattering Cross SectionLet us consider only those interactions that scatter the particles by u ¼908or more.For hard spheres,taking the angle of incidence equal to the angle of reflection,the 908collision occurs on the x ¼458diagonal (see Fig.3.2),therefore having a cross section s 90¼p a 2122,(3:1:11)which is a factor of two smaller than (3.1.4).Of course,multiple collisions at smaller angles (radii larger than a 12=ffiffiffi2p )also eventually scatter incident particles through 908.This indeterminacy indicates that a more precise way of determining the scat-tering cross section is required.For this purpose we introduce a differential scatter-ing cross section I (v ,u ).Consider a beam of particles incident on a scattering center (again assumed fixed),as shown in Figure 3.3.We assume that the scattering force is symmetric about the line joining the centers of the two particles.A particle incident at a distance b off-center from the target particle is scattered through an angle u ,as shown in Figure 3.3.The quantity b is the impact parameter and u is the scattering angle (see also Fig.3.2).Now,flux conservation requires that for incoming flux G ,G 2p b d b ¼ÀG I (v ,u )2p sin u d u (3:1:12)FIGURE 3.2.Hard-sphere scattering.46ATOMIC COLLISIONS3.1BASIC CONCEPTS47FIGURE3.3.Definition of the differential scattering cross section.that is,that all particles entering through the differential annulus2p b d b leave through a differential solid angle d V¼2p sin u d u.The minus sign is because an increase in b leads to a decrease in u.The proportionality constant is just I(v,u), which has the dimensions of area per steradian.From(3.1.12)we obtainI(v,u)¼bsin ud bd u(3:1:13)The quantity d b=d u is determined from the scattering force,and the absolute value is used since d b=d u is negative.We will calculate I(v,u)for various potentials in Section3.2.We can calculate the total scattering cross section s sc by integrating I over the solid angles sc¼2p ðpI(v,u)sin u d u(3:1:14)It is clear that s sc¼s for scattering through any angle,as defined in(3.1.2).It is often useful to define a different cross sections m¼2p ðp(1Àcos u)I(v,u)sin u d u(3:1:15)The factor(1Àcos u)is the fraction of the initial momentum m v lost by the incident particle,and thus(3.1.15)is the momentum transfer cross section.It is s m that is appropriate for calculating the frictional drag in the force equation(2.3.9).For asingle velocity,we would just have n m¼s m v,where s m is generally a function of velocity.In the macroscopic force equation(2.3.15),n m must be obtained by aver-aging over the particle velocity distributions,which we do in Section3.5.We illustrate the use of the differential scattering cross section to calculate thetotal scattering and momentum transfer cross sections for the hard-sphere modelshown in Figure3.2.The impact parameter is b¼a12sin x,and differentiating, d b¼a12cos x d x,so thatb d b¼a212sin x cos x d x¼12a212sin2x d x(3:1:16)From Figure3.2the scattering angle u¼pÀ2x,such that(3.1.16)can be written asb d b¼À1a212sin u d u(3:1:17)48ATOMIC COLLISIONSSubstituting(3.1.17)into(3.1.13),we haveI(v,u)¼14a212(3:1:18)Using the definitions of s sc and s m in(3.1.14)and(3.1.15),respectively,wefinds sc¼s m¼p a212(3:1:19) for hard-sphere collisions.In general,s sc=s m for other scattering forces.For electron collisions with atoms the electron radius is negligible compared to the atomic radius so that a12%a,the atomic radius.Although the value of a% 10À8cm gives s sc¼s m%3Â10À16cm2,which is reasonable,it does not capture the scaling of the cross section with speed.In the following sections of this chapter,we consider collisional processes in more detail.Except for Coulomb collisions,we confine our attention to electron–atom and ion–atom processes.After a discussion of collision dynamics in Section3.2,we describe elastic collisions in Section3.3and inelastic collisions in Section3.4.We reserve a discussion of some aspects of inelastic collisions until Chapter8,in which a more complete range of atomic and molecular processes is considered.In Section3.5,we describe the averaging over particle velocity distri-butions that must be done to obtain the collisional rate constants.Experimental values for argon are also given in Section3.5;these are needed for discussing energy transfer and diffusive processes in the succeeding chapters.A more detailed account of collisional processes,together with many results of experimental measurements,can be found in McDaniel(1989),McDaniel et al.(1993),Massey et al.(1969–1974),Smirnov(1981),and Raizer(1991).3.2COLLISION DYNAMICSCenter-of-Mass CoordinatesIn a collision between projectile and target particles there is recoil of the target as well as deflection of the projectile.In fact,both may be moving,and,in the case of like-particle collisions,not distinguishable.To describe this more complicated state,a center-of-mass(CM)coordinate system can be introduced in which projec-tiles and targets are treated equally.Without loss of generality,we can transform to a coordinate system in which one of the particles is stationary before the collision. Hence,we consider a general collision in the laboratory frame between two particles having mass m1and m2,position r1and r2,velocity v1and v2;0,and scattering angle u1and u2,as shown in Figure3.4a.We assume that the force F acts along the line joining the centers of the particles,with F12¼ÀF21.3.2COLLISION DYNAMICS49The center-of-mass coordinates may be defined by the linear transformationR ¼m 1r 1þm 2r 2m 1þm 2(3:2:1)andr ¼r 1Àr 2(3:2:2)with the accompanying CM velocityV ¼m 1v 1þm 2v 2m 1þm 2(3:2:3)and the relative velocityv R ¼v 1Àv 2(3:2:4)v 2´m 1m R center(a )(b )FIGURE 3.4.The relation between the scattering angles in (a )the laboratory system and (b )the center-of-mass (CM)system.50ATOMIC COLLISIONSThe force equations for the two particles are:m1_v1¼F12(r),m2_v2¼F21(r)¼ÀF12(r)(3:2:5) Adding these equations we get the result for the CM motion that_V¼0,such that the CM moves with constant velocity throughout the collision.Now dividing thefirst of (3.2.5)by m1and the second by m2,and using the definition in(3.2.4)we havem R_v R¼F12(r)(3:2:6) which is the equation of motion of a“fictitious”particle with a reduced massm R¼m1m2m1þm2(3:2:7)in afixed central force F12(r).Thefictitious particle has mass m R,position r(t), velocity v R(t),and scattering angle Q,as shown in Figure3.4b.This result holds for any central force,including the hard-sphere,Coulomb,and polarization forces that we subsequently consider.If(3.2.6)can be solved to obtain the motion,includ-ing Q,then we can transform back to the laboratory frame to get the actual scattering angles u1and u2.It is easy to show from momentum conservation(Problem3.2)thattan u1¼sin Q(m1=m2)(v R=v0R)þcos Q(3:2:8a)andtan u2¼sin Qv R=v0RÀcos Q(3:2:8b)where v R and v0R are the speeds in the CM system before and after the collision, respectively.For an elastic collision,the scattering force can be written as the gradient of a potential that vanishes as r¼j r j!1:F12¼Àr U(r)(3:2:9) It follows that the kinetic energy of the particle is conserved for the collision in the CM system.Hence v0R¼v R,and we obtain from(3.2.8)thattan u1¼sin Q1=m2þcos Q(3:2:10)3.2COLLISION DYNAMICS51and,using the double-angle formula for the tangent,u2¼1(pÀQ)(3:2:11) For electron collisions with ions or neutrals,m1=m2(1and we obtain m R%m1 and u1%Q.For collision of a particle with an equal mass target,m1¼m2,we obtain m R¼m1=2and u1¼Q=2.Hence for hard-sphere elastic collisions against an initially stationary equal mass target,the maximum scattering angle is908.Since the same particles are scattered into the differential solid angle 2p sin Q d Q in the CM system as are scattered into the corresponding solid angle 2p sin u1d u1in the laboratory system,the differential scattering cross sections are related byI(v R,Q)2p sin Q d Q¼I(v R,u1)2p sin u1d u1(3:2:12)where d Q=d u1can be found by differentiating(3.2.10).Energy TransferElastic collisions can be an important energy transfer process in gas discharges,and can also be important for understanding inelastic collision processes such as ioniz-ation,as we will see in Section3.4.For the elastic collision of a projectile of mass m1 and velocity v1with a stationary target of mass m2,the conservation of momentum along and perpendicular to v1and the conservation of energy can be written in the laboratory system asm1v1¼m1v01cos u1þm2v02cos u2(3:2:13)0¼m1v01sin u1Àm2v02sin u2(3:2:14)1 2m1v21¼12m1v012þ12m2v022(3:2:15)where the primes denote the values after the collision.We can eliminate v01and u1 and solve(3.2.13)–(3.2.15)to obtain1 2m2v022¼12m1v214m1m2(m1þm2)2cos2u2(3:2:16)Since the initial energy of the projectile is12m1v21and the energy gained bythe target is12m2v022,the fraction of energy lost by the projectile in the laboratory52ATOMIC COLLISIONSsystem isz L¼4m1m2(m12)cos2u2(3:2:17) Using(3.2.11)in(3.2.17),we obtainz L¼2m1m2(m1þm2)2(1Àcos Q)(3:2:18)where Q is the scattering angle in the CM system.We average over the differential scattering cross section to obtain the average loss:k z L l Q¼2m1m2(m1þm2)2Ð(1Àcos Q)I(v R,Q)2p sin Q d Q ÐI(v R,Q)2p sin Q d Q¼2m1m2 (m1þm2)2s ms sc(3:2:19)where s sc and s m are defined in(3.1.14)and(3.1.15).For hard-sphere scattering of electrons against atoms,we have m1¼m(electron mass)and m2¼M(atom mass),and s sc¼s m by(3.1.19),such that k z L l Q¼2m=M 10À4.Hence electrons transfer little energy due to elastic collisions with heavy particles,allowing T e)T i in a typical discharge.On the other hand,for m1¼m2,we obtain k z L l Q¼12,leading to strong elastic energy exchange among heavy particles and hence to a common temperature.Small Angle ScatteringIn the general case,(3.2.6)must be solved to determine the CM trajectory and the scattering angle Q.We outline this approach and give some results in Appendix A. Here we restrict attention to small-angle scattering(Q(1)for which the fictitious particle moves with uniform velocity v R along a trajectory that is practi-cally unaltered from a straight line.In this case,we can calculate the transverse momentum impulse D p?delivered to the particle as it passes the center of force at r¼0and use this to determine Q.For a straight-line trajectory,as shown in Figure3.5,the particle distance from the center of force isr¼(b2þv2R t2)1=2(3:2:20)where b is the impact parameter and t is the time.We assume a central force of the form(3.2.9)withU(r)¼C(3:2:21)3.2COLLISION DYNAMICS53where i is an integer.The component of the force acting on the particle perpendicu-lar to the trajectory is (b =r )j d U =d r j .Hence the momentum impulse isD p ?¼ð1À1b r d U d r d t (3:2:22)Differentiating (3.2.20)to obtaind t ¼r v R d r(r 2Àb 2)1=2substituting into (3.2.22),and dividing by the incident momentum p k ¼m R v R ,we obtainQ ¼D p ?p k ¼2b m R v R ð1b d U d r d r (r 22)(3:2:23)The integral in (3.2.23)can be evaluated in closed form (Smirnov,1981,p.384)to obtainQ ¼AW R b (3:2:24)where W R ¼12m R v 2R is the CM energy andA ¼C ffiffiffiffip p G ½(i þ1)=2 (3:2:25)FIGURE 3.5.Calculation of the differential scattering cross section for small-angle scattering.The center-of-mass trajectory is practically a straight line.54ATOMIC COLLISIONSwith G ,the Gamma function.ÃInverting (3.2.24),we obtainb ¼A W R Q1=i (3:2:26)and differentiating,we obtaind b ¼À1i A W R 1=i d Q Q (3:2:27)Substituting (3.2.26)and (3.2.27)into (3.1.13),with sin Q %Q ,we obtain the differ-ential scattering cross section for small angles:I (v R ,Q )¼1i A W R 2=i 1Q 2þ2=i (3:2:28)The variation of s ,n ,and K with v R are determined from (3.2.28)and the basic definitions in Section 3.1.If (3.2.28)is substituted into (3.1.14)or (3.1.15),then we see that a scattering potential U /r Ài leads to s /v À4=i R and n /K /v À(4=i )þ1R .These scalings are summarized in Table 3.1for the important scattering processes,which we describe in the next section.3.3ELASTIC SCATTERINGCoulomb CollisionsThe most straightforward elastic scattering process is a Coulomb collision between two charged particles q 1and q 2,representing an electron–electron,electron–ion,or ion–ion collision.The Coulomb potential is U (r )¼q 1q 2=4pe 0r such that i ¼1and TABLE 3.1.Scaling of Cross Section s ,Interaction Frequency n ,and Rate Constant K ,With Relative Velocity v R ,for VariousScattering Potentials UProcessU (r )s n or K Coulomb1/r 1/v R 41/v R 3Permanent dipole1/r 21/v R 21/v R Induced dipole1/r 41/v RConst Hard sphere 1/r i ,i !1Const v RÃG (l )¼(l À1)!¼l G (l À1)with G (1=2)¼ffiffiffiffip p .3.3ELASTIC SCATTERING 55we obtainA¼C¼q1q2 4pe0from(3.2.25).Using this in(3.2.28),wefindI¼b0Q2(3:3:1)whereb0¼q1q240W R(3:3:2)is called the classical distance of closest approach.The differential scattering cross section can also be calculated exactly,which we do in Appendix A,obtaining the resultI¼b04sin(Q=2)2(3:3:3)However,due to the long range of the Coulomb forces,the integration of I oversmall Q(large b)leads to an infinite scattering cross section and to an infinitemomentum transfer cross section,such that an upper bound to b,b max,must beassigned.This is done by setting b max¼l De,the Debye shielding distance for a charge immersed in a plasma,which we calculated in Section2.4.For momentumtransfer,the dependence of s m on l De is logarithmic(Problem3.5),and the exact choice of b max(or Q min)makes little difference.For scattering,s sc pl2De, which is a very large cross section that depends sensitively on the choice of b max. However,we are generally not interested in scattering through very small angles, which do not appreciably affect the discharge properties.The cross section for scattering through a large angle,say Q!p=2,is of more interest.There are two processes that lead to a large scattering angle Q for a Coulombcollision:(1)a single collision scatters the particle by a large angle;(2)the cumu-lative effect of many small-angle collisions scatters the particle by a large angle.Thetwo processes are illustrated in Figure3.6;the latter process is diffusive and,as wewill see,dominates the former.To estimate the cross section s90(sgl)for a single large-angle collision,we inte-grate(3.3.3)over solid angles from p=2to p to obtain(Problem3.6)s90(sgl)¼14p b2(3:3:4)To estimate s90(cum)for the cumulative effect of many collisions to produce a p=2deflection,wefirst determine the mean square scattering angle k Q2l1for a 56ATOMIC COLLISIONSsingle collision by averaging Q 2over all permitted impact parameters.Since the col-lisions are predominantly small angle for Coulomb collisions,we can use (3.2.24),which is Q ¼b 0=b .Hencek Q 2l 1¼1p b 2max ðb max b min q 1q 24pe 0W R 22p b d b b 2(3:3:5)The integration has a logarithmic singularity at both b ¼0and b ¼1,which is cut off by the finite limits.The singularity at the lower limit is due to the small-angle approximation.Setting b min ¼b 0=2is found to approximate a more accurate calcu-lation.The upper limit,as already mentioned,is b max ¼l De .Using these values and integrating,we obtaink Q 2l 1¼2p b 20p b 2max ln L (3:3:6)where L ¼2l De =b 0)1.The number of collisions per second,each having a cross section of p b 2max orsmaller,is n g p b 2max v R ,where n g is the target particle density.Since the spreadingof the angle is diffusive,we can then writek Q 2l (t )¼k Q 2l 1n g p b 2max v R tSetting t ¼t 90at k Q 2l ¼(p =2)2and using (3.3.6),we obtain (see also Spitzer,1956,Chapter 5)n 90¼t À190¼n g v R 8p b 20lnLFIGURE 3.6.The processes that lead to large-angle Coulomb scattering:(a )single large-angle event;(b )cumulative effect of many small-angle events.3.3ELASTIC SCATTERING 57Writing n90¼n g s90v R,we see thats90¼8p b 2ln L(3:3:7)Although L is a large number,typically ln L%10for the types of plasmas we are considering.Comparing s90(sgl)to s90,we see that due to the large range of the Coulomb fields,the effective cross section for many small-angle collisions to produce a root mean square(rms)deflection of p=2is larger by a factor(32=p2)ln L. Because of this enhancement,it is possible for electron–ion or ion–ion particle col-lisions to play a role in weakly ionized plasmas(say one percent ionized).Another important characteristic of Coulomb collisions is the strong velocity dependence. From(3.3.2)we see that b0/1=v2R.Thus,from(3.3.4)or(3.3.7)s90/1v4R(3:3:8)such that low-velocity particles are preferentially scattered.The temperature of the species is therefore important in determining the relative importance of the various species in the collisional processes,as we shall see in subsequent sections.Polarization ScatteringThe main collisional processes in a weakly ionized plasma are between charged and neutral particles.For electrons at low energy and for ions scattering against neutrals, the dominant process is relatively short-range polarization scattering.At higher energies for electrons,the collision time is shorter and the atoms do not have time to polarize.In this case the scattering becomes more Coulomb-like,but with b max at an atomic radius,inelastic processes such as ionization become important as well.The condition for polarization scattering is v R.v at,where v at is the charac-teristic electron velocity in the atom,which we obtain in the next section.Because of the short range of the polarization potential,we need not be concerned with an upper limit for the integration over b,but the potential is more complicated.We determine the potential from a simple model of the atom as a point charge of valueþq0,sur-rounded by a uniform negative charge sphere(valence electrons)of total chargeÀq0,such that the charge density is r¼Àq0=43p a3,where a is the atomic radius.An incoming electron(or ion)can polarize the atom by repelling(or attracting) the charge cloud quasistatically.The balance of forces on the central point charge due to the displaced charge cloud and the incoming charged particle,taken to have charge q,is shown in Figure3.7,where the center of the charge cloud and the point charge are displaced by a distance d.Applying Gauss’law to a sphere 58ATOMIC COLLISIONSof radius d around the center of the cloud,4pe0d2E ind¼Àq0d3 awe obtain the induced electricfield acting on the point charge due to the displaced cloudE ind¼Àq0d 4pe0a3The electricfield acting on the point charge due to the incoming charge isE appl¼q 4pe0rFor force balance on the point charge,the sum of thefields must vanish,yielding an induced dipole moment for the atom:p d¼q0d¼qa3r2(3:3:9)The induced dipole,in turn,exerts a force on the incoming charged particle:F¼2p d q4pe0r3^r¼2q2a34pe0r5^r(3:3:10)FIGURE3.7.Polarization of an atom by a point charge q.3.3ELASTIC SCATTERING59Integrating F with respect to r,we obtain the attractive potential energy:U(r)¼Àq2a38pe0r4(3:3:11)The polarizability for this simple atomic model is defined as a p¼a3.The relative polarizabilities a R¼a p=a30,where a0is the Bohr radius,for some simple atoms and molecules are given in Table3.2.The orbits for scattering in the polarization potential are complicated(McDaniel, 1989).As shown in Figure3.8,there are two types of orbits.For impact parameter b.b L,the orbit has a hyperbolic character,and for b)b L,the straight-line trajec-tory analysis in Section3.2can be applied(Problem3.7).For b,b L,the incoming particle is“captured”and the orbit spirals into the core,leading to a large scattering angle.Either the incoming particle is“reflected”by the core and spirals out again,or the two particles strongly interact,leading to inelastic changes of state.The critical impact parameter b L can be determined from the conservation of energy and angular momentum for the incoming particle having mass m and speed v0,with the mass of the scatterer taken to be infinite for ease of analysis.In cylindrical coordinates(see Fig.3.8a),we obtain1 2m v2¼12m(_r2þr2_f2)þU(r)(3:3:12a)m v0b¼mr2_f(3:3:12b)TABLE3.2.Relative Polarizabilities a R5a p/a03ofSome Atoms and Molecules,Where a0is the Bohr RadiusAtom or Molecule a RH 4.5C12.N7.5O 5.4Ar11.08CCl469.CF419.CO13.2CO217.5Cl231.H2O9.8NH314.8O210.6SF630.Source:Smirnov(1981).60ATOMIC COLLISIONSAt closest approach,_r¼0and r ¼r min .Substituting these into (3.3.12)and elimi-nating _f ,we obtain a quadratic equation for r 2min:v 20r 4min Àv 20b 2r 2min þa p q 240m¼0Using the quadratic formula to obtain the solution for r 2min ,we see that there is noreal solution for r 2min when(v 20b 2)2À4v 20a p q 20 0Choosing the equality at b ¼b L ,we solve for b L to obtains L ¼p b 2L ¼pa p q 2e 0 1=21v 0(3:3:13)which is known as the Langevin or capture cross section.If the target particle has a finite mass m 2and velocity v 2and the incoming particle has a mass m 1and velocity v 1,then (3.3.13)holds provided m is replaced by the reduced mass m R ¼m 1m 2=(m 1þm 2)and v 0is replaced by the relative velocity v R ¼j v 1Àv 2j .We (a )(b )FIGURE 3.8.Scattering in the polarization potential,showing (a )hyperbolic and (b )captured orbits.3.3ELASTIC SCATTERING 61。

从都灵到老外滩意大利都灵工业设计展点燃宁波创意火花佚名
【期刊名称】《流行色》
【年(卷),期】2007(000)002
【摘要】无
【总页数】2页(P154-155)
【正文语种】中文
【相关文献】
1.意大利学系列篇(三)都灵理工大学工业设计专业教学纪实 [J], 吴佩平
2.“工具箱”都灵办公实验室,都灵,意大利 [J], 李若星
3.SER.MI.G.:前都灵军械厂及周边建筑改造,都灵,意大利 [J], 司马蕾
4.中国美术学院设计艺术学院工业设计系意大利都灵理工大学设计艺术学院2009合作workshop—vending machine课程教学交流 [J], 吴佩平
5.意大利工业设计展点燃宁波创意火花 [J],
因版权原因，仅展示原文概要，查看原文内容请购买。

a r X i v :c o n d -m a t /0201141v 1 [c o n d -m a t .m t r l -s c i ] 9 J a n 2002First-Principles-Based Thermodynamic Description of SolidCopper Using the Tight-Binding ApproachSven P.Rudin,1M.D.Jones,2C.W.Greeff,1and R.C.Albers 11Los Alamos National Laboratory,Los Alamos,NM 875452Department of Physics and Center for Computational Research,University at Buffalo,The State University of New York,Buffalo,NY 14260(February 1,2008)Abstract A tight-binding model is fit to first-principles calculations for copper that include structures distorted according to elastic constants and high-symmetry phonon modes.With the resulting model the first-principles-based phonon dispersion and the free energy are calculated in the quasi-harmonic approximation.The resulting thermal expansion,the temperature-and volume-dependence of the elastic constants,the Debye temperature,and the Gr¨u neisen parameter are compared with available experimental data.PACS numbers:63.20.Dj,64.30.+t,65.40.De,65.40.GrTypeset using REVT E XI.INTRODUCTIONDensity-functional theory(DFT)first-principles electronic-structure methods describe anomaly-free solids such as elemental copper successfully.They achieve high accuracy for quantities such as bulk properties,1surface relaxation and lattice dynamics of the surface,2 as well as the epitaxial Bain path and elastic constants.3DFT methods are routinely used to compute the zero-temperature internal energy,Φ0(V),but also can be used to calculate the free energy contributions from the ions,F I(V,T),and the electrons,F E(V,T),resulting in a complete equation of state,F(V,T)=Φ0(V)+F I(V,T)+F E(V,T).(1)However,the required computational effort is expensive,and an alternative efficient evalu-ation at all volumes and temperatures would be desirable.In this paper we use the computationally less demanding tight-binding(TB)total energy model in conjunction with well chosenfirst-principles calculations.In particular,we use the functionalfitting forms developed at the U.S.Naval Research Laboratory(NRL)for com-puting the total energy within the TB formalism,i.e.,without an external potential.4The model isfit to and accurately reproduces a set offirst-principles calculations with a speed-up of many orders of magnitude.In addition,transferability(i.e.,a TB parameterization that is accurate for a wide variety of crystal structures and atomic arrangements)has been successfully demonstrated for semiconductors as well as for simple and transition metals.4 We believe that the TB method can be used as a highly accurate,but computationally more efficient,surrogate for a fullfirst-principles-based approach to calculate the equation of state for solids.Copper is frequently used as a test material for theoretical methods.5In this paper we have(1)developed an improvedfit for copper that is accurate for phonons,and(2)used this model to calculate a wide range of temperature-and volume-dependent thermodynamic quantities.Copper is furthermore widely employed as a pressure standard in high-pressure research.7 This use is based on correcting P(V)data taken along the shock Hugoniot8to room tem-perature.Such corrections employ model assumptions about the volume dependence of the Gr¨u neisen parameterγ(V),which is difficult to measure independently.Shock heating increases with pressure,making the corrections more significant at high pressure.It is there-fore important to develop theoretical techniques for accurate prediction ofγfor copper at high pressure.Phonons play a major role in the calculations of thermodynamic quantities,and the TBfits are adjusted to more accurately calculate them.Structures corresponding to high-symmetry phonon modes are shown here to aid in refining the model;the resulting phonon density of states can then be used to determine the free energy and hence all thermodynamic quantities of interest.The precision required to calculate phonon frequencies is an order of magnitude higher than that for the lattice constant or bulk modulus,9making this a stringent test for the validity of the tight-binding approach in general and the copper model in particular.The ion–ion free energy of Eq.1is often separated into harmonic and anharmonic parts,F I(V,T)=F H(V,T)+F A(V,T).(2)Normally,the harmonic component is not a function of volume,but is calculated from the effect of small displacements about the zero-temperature equilibrium lattice.In our calcula-tions,we use the quasi-harmonic approximation,which considers small displacements at any fixed volume(lattice constant)within the harmonic approximation,and hence our phonon frequencies become volume dependent.However,our phonon frequencies are calculated at zero temperature for any given volume,and are not temperature dependent.The anharmonic part of the free energy involves terms that arise from the potential energy of the lattice when it is expanded beyond the harmonic part to higher than second order. Such terms are needed at high temperatures,when the phonon amplitudes are large,and ultimately lead to melting.They are also needed to explain thermal expansion effects whenthe harmonic part is based on the equilibrium volume.The quasi-harmonic approximation can handle thermal expansion and the Gr¨u neisen parameter accurately through the volume dependence of the phonons at low temperature.At sufficiently high temperatures,the quasi-harmonic approximation breaks down when the phonon amplitudes become large,and additional anharmonic phonon-phonon corrections are necessary(as indicated in Eq.2).We have not included these anharmonic types of effects in our calculations.Hence we always set F A(V,T)=0,and our calculations will become less reliable at very high temperatures (near melting).In the following section we introduce the basic ideas of the tight-binding method and the first-principles method used to generate thefitting database,and then describe our TBfitting procedures.In the subsequent section we present calculated results for the thermodynamic properties and compare them with experiment.II.FITTING THE MODELA.Tight-binding electronic structureThe tight-binding approach is essentially a parameterized version of thefirst-principles calculations and hence is orders of magnitude more computationally efficient.In DFT meth-ods the secular equation,Hψi,v=ǫi,v Sψi,v,(3)is constructed directly from approximate solutions to the full many-body Hamiltonian,and involves a self-consistent potential that is solved iteratively;whereas in the TB approach the elements of the Hamiltonian(and the overlap matrix)themselves have been parameter-ized.Only two-center terms are considered.10For the non-orthogonal tight-binding model described here this requires73fitted parameters.Of those parameters,thirty each are used to describe the inter-site matrix elements of the Hamiltonian and of the overlap matrix.For each combination of symmetries(ll′m)theform11ish ll′m(r)=(a ll′m+b ll′m r)e−c2ll′m r f c(r),(4)s ll′m(r)= ¯a ll′m+¯b ll′m r e−¯c2ll′m r f c(r),(5) where f c=1/(1+e2(r−r0))is a multiplicative factor included to ensure a smooth cutoffwithincreasing distance.In our calculations we have set r0=16.0Bohr radii.The remaining13parameters determine the on-site terms,which allows the parameter-ization to be applied to structures not included in thefitting database.A measure of thevalence electron density,ρ= i=j e−λ2r ij f c(r ij),(6) where r ij is the interatomic distance,serves to describe the on-site energy,eα=e0α+e1αρ2/3+e2αρ4/3+e3αρ2,(7)for the three orbital typesα,i.e.,s,p,and d.These terms are somewhat similar to an embedded-atom-like form in that the energy changes depending on the nearby arrangements of atoms,and may approximately account for self-consistency effects as the atoms move around.B.Full potential LAPW methodThefirst-principles quality of the tight-binding model results fromfitting to full potential linear augmented plane wave(LAPW)calculations using the reliable WIEN97program suite.12The parameters for thefirst-principles calculations are listed in Table I.The LAPW method divides space into spherical regions centered on the atoms and the remaining interstitial region.The radius of the spheres,the muffin-tin radius R m,must be chosen such that the spheres do not overlap.The basis functions used to represent the wave function are adapted to the regions:radial solutions to the Schr¨o dinger equation in thespheres,plane waves in the interstitial region.The wave functions then are found iteratively within density-functional theory,constrained to match at the boundaries of the different regions.C.Initial Fitting ProcedureWefirstfit the TB method to predict energy differences between the ground-state and non-equilibrium structures.Thefitting database includedfirst-principles energies calculated for the cubic structures.In addition to the total energies of these structures,it proved to be crucial tofit the energy bands at high-symmetry points in reciprocal space.15,16By decom-posing the electronic wave function in terms of the symmetry character of the eigenvalues17 the bands are guided to the correct ordering.The total energies and the band energies can be calculated by starting with a very crude initial tight-binding model that ignores intersite terms;15the errors are then minimized utilizing standard nonlinear least squares algorithms.18Figure1shows the T=0phonon dispersion for fcc copper calculated with the initial model.15The long-wavelength modes nearΓare well described,the short-wavelength modes near the zone boundary display somewhat high frequencies,in particular the longitudinal modes.The reasonable agreement for phonons near the zone centerΓcan be understood by considering the elements of thefitting database.The bulk modulus,i.e.,a linear combination of the elastic constants,is implicitly included in thefit.While this does not guarantee accurate elastic constants,i.e.,good agreement for the slopes of the dispersion nearΓ,it does set the right scale.Furthermore,thefit includes the bcc structure,which is related to the fcc crystal by a tetragonal strain corresponding to the long-wave-vector limit of the longitudinal mode in the[00ξ]direction.The database lacks any information related to the short-wave-vector modes.D.Fitting procedure with distorted structuresIn order to construct a model with an improved phonon dispersion the database was expanded to include additional information on the phonons,in particular,structures that are snapshots of the crystal deformed by particular phonon modes,i.e.,frozen phonons. The undistorted and distorted crystal structures are treated on the same footing in the first-principles calculations and thefitting procedure,implicitly including the differences in energy and hence the frequencies of the phonon modes.The longitudinal and the transverse mode at the high-symmetry point X(q=(0,0,1)) were chosen because of the large discrepancy in frequency(see Fig.1)and because the distorted structures require only a doubling of the unit cell.These distorted structures are considered as additional,distinct structures in the database,to befit to over a range of volumes.The initialfit for copper already contains some of the character of distortions related to the elastic constants:the bulk modulus is explicitly included in the energy as a function of volume,and the tetragonal distortion of the fcc crystal is somewhat reflected byfitting to the bcc structure.For completeness,tetragonally-and trigonally-distorted fcc crystals were added to thefit as distinct structures.These additional structures barely influence the model resulting from thefit;however,thefitting process converges much more quickly when they are included.The cubic structures that were included in the initialfit differ from each other by an energy scale of fractions of electron volts.Phonons require a model tuned to discern energies on a scale that is approximately an order of magnitude smaller.This could be a problem since the minimization procedure tends to ignore small energy differences.For frozen phonons at the zone boundary,where neighboring atoms move against each other,it turns out that amplitudes which are still within the harmonic regime can produce energies that differ from the undistorted structure by fractions of electron volts.The distortions corresponding to elastic constants,however,need to be exaggerated for them to give large enough energydifferences.The trigonal distortion used here compresses the base angle from90◦to75◦, while the tetragonal distortion changes the c/a ratio from unity to1.9.Figure2shows the energy values in thefitting database alongside those of the initial and improved tight-binding models.The volumes of thefirst-principles calculations are limited to structures where the muffin-tin radius R m is smaller than the nearest-neighbor distance, particularly for the strongly-distorted fcc structures the choice of R m=2.0a.u.prohibits strong compression.No such limitations exist for the tight-binding approach;the volumes for which the model is appropriate will become clear in the next section.Figure3shows the errors in the improved model’sfipared to the initialfit,errors for the simple,cubic structures remain about the same.The errors for the tetragonally-distorted structures are small around the equilibrium volume(11.93˚A3),but show a ten-dency to increase as the crystal is compressed.The form of the matrix elements(Eq.4) cannot be expected to allow a high-qualityfit at all volumes;indeed when only a subset of data points are included in thefit the errors show no radical change.Including the distorted structures in thefit improves the transferability of the model. Figure4shows the improved agreement between tight-binding andfirst-principles energies for the diamond structure,which is not included in thefit.The transferability to a structure of such a different coordination is not guaranteed,and our initial model did not reproduce the diamond energies well,nor did the model of the NRL group.5Figure5shows the phonon dispersion calculated with the improved model.Including the distorted fcc structures clearly refines the agreement with the measured values,though the curves do not overlap perfectly:the dispersion of the low-lying transverse modes in the[0ξ1] direction shows a different character,and the high-frequency longitudinal modes remain somewhat large.The discrepancy of the longitudinal frequency at L suggests including this data point in thefit.However,afirst-principles,frozen-phonon calculation of this mode shows better agreement with the tight-binding model than with experiment and was therefore not added to the database.Figure6shows the phonon density of states calculated with the improved model.Thegeneral shape agrees with the data calculated from the Born-von K´a rm´a n force constants fitted to the experimental phonon dispersion along high-symmetry directions.19,20The differ-ence in maximum frequencies and the peak near7THz can be attributed to the discrepancy in the dispersion of the longitudinal mode near L in the[ξξξ]direction.The tight-binding density of states displays more structure around4THz,which may be due to modes in low-symmetry directions that are not part of the experimental force-constant model.The distorted structures added to thefit indeed make for a model that is better suited for phonon calculations.However,while the additional constraints improve the total energies described by the model,the electronic band structure deteriorates.Figure7shows the electronic band structure along two sample high-symmetry directions of fcc copper at the experimental volume.While the initial model agrees well with thefirst-principles band structure,the model improved for thermodynamic quantities loses the good agreement. The resulting electronic density of states,shown in Fig.8,shows the same discrepancy; however,the density of states at the Fermi energy is quite similar,which is important for the temperature-dependent influence of the electrons(see below).It is possible that a better or moreflexible functional form for the distance dependence of the intersite Hamiltonian and overlap matrices are necessary to keep the good transferability and the good agreement with the individual energy bands.III.CALCULATIONS WITH THE TB MODELA.Force ConstantsThe force constants are calculated from the tight-binding model by the direct-force method,21–24which relies on evaluating the forces on all atoms in a simulation cell in which a reference atom(0,i)has been displaced.The large simulation cell consists of primitive cells transposed by vectorsℓ.Due to periodic boundary conditions on the simulation cell, the force on an atom(ℓ,j)is in response to the displaced reference atom(0,i)as well as itsimages transposed by vectors L,F(ℓ,j)=− LφC(ℓ,j;0,i)= Lφ∂uα(0,i)≈−Fβ(ℓ,j)(q),which in turn is the Fourier transform of the system’s force constants,Dαβ(q)=1that break the inversion symmetry with respect to the reference atom have to be duplicated (with adjusted weight)and transposed with a basis vector of the simulation cell to reinstate the symmetry.The cubic symmetry of the fcc crystal allows the calculation of the force constants at a particular volume with a single displacement of the basis atom.Distorted fcc structures no longer have the cubic symmetry,the calculation of the force constants therefore requires the forces to be evaluated for the basis atom displaced in all three Cartesian directions separately. For all calculations the simulation cell contained108atoms and a mesh of4×4×4k-points was used.B.ThermodynamicsAs indicated by Eq.1,the free energy is the internal energy from the tight-binding calculation with entropic terms added from the electrons and the ions.In both terms the relevant physical quantity is the density of states(DOS).The electronic DOS,n(E),the occupation of which is given by the Fermi distribution f(E,T)=[e(E−E f)/(k B T)+1]−1, determines the electrons’contribution to the entropy,S el(T)=−k B [f ln f+(1−f)ln(1−f)]n(E)dE.(12) The phonon DOS,g(ω),contributes through the zero-point energy,1U zero=the phonons,although at low temperatures(where both contributions are very small)and small volumes the percentage rises to about10%.Figure9shows the resulting free-energy as a function of volume for temperatures between 0K to1400K(at ambient pressure copper melts at1356K;melting is an anharmonic effect that lies outside the scope of the quasi-harmonic treatment)in100K increments.A comparison with the free energy for the bcc phase shows the fcc structure at lower free energy for all temperatures and volumes,indicating that the model agrees with experiment in that respect.The free energy as a function of volume and temperature determines the thermal ex-pansion.The temperature-dependent lattice constant derived from the tight-binding model is shown in the inset of Fig.9along with the experimental values.As is typical for GGA-calculations,the tight-binding model overestimates the equilibrium volume by1.4%.The calculated linear expansion coefficient is compared to experimental data in Fig.10and shows good agreement,in particular the characteristic temperature,which is determined by the phonon characteristic temperatures(see below).The shape of the free energy as a function of volume and temperature directly provides the temperature-dependence of the bulk modulus,B(T),which is calculated byfitting a second order Birch equation of state.26The bulk modulus is related to two of the elastic constants by B(T)=1accounted for byfinding the equilibrium volume for each temperature and then calculating the effect of the strain on the free energy of that volume.Figure12shows the calculated T=0elastic constants as a function of volume.The phonon characteristic temperatures,which are defined as moments of the phonon density of states,28ln(k Bθ0)= ln(¯hω) BZ,(15)k Bθ1=43(¯hω)2 BZ 1/2,(17) are shown in Fig.13.The approximate rule of thumbθ2≈θ1≈e1/3θ0holds nicely for the calculated values(inset).At temperatures below the phonon characteristic temperatures individual phonon modes must be considered separately,because they contribute to the crystal’s thermal properties with weights depending on their frequency relative to the temperature.The weight of a mode of branch s with wave vector q is determined by the heat capacity for that mode,c s(q)=∂eβ¯hωs(q)−1.(18)The sum of these individual heat capacities as a function of temperature agrees well with calorimetric data;the comparison is plotted in Fig.14in terms of the Debye temperature θD,which is found such that the Debye model’s heat capacityc V=9k B T(e x−1)2dx(19) is the same as the heat capacity calculated for the tight-binding model at the same temper-ature.The shape of the Debye temperature plotted against temperature remains very similar with compression;the curve itself is shifted upwards with the same volume-dependence as the characteristic phonon temperatures.The heat capacity of each individual phonon mode,combined with the Gr¨u neisen Pa-rameter of that mode,d lnωs(q)γq,s=−q,s c v,s(q).(21) At high temperatures(T>θ2),where all phonon modes contribute equally,γ≈γ0= d lnθ0/d lnρ.At low temperatures only the acoustic phonon modes contribute.Figure15compares the tight-binding results for the Gr¨u neisen parameter with available data.For densities up to near13g/cm3the results roughly agree with the rule of thumb thatγ·ρ=constant.Our values are slightly below the experimental values,indicating that the phonon frequencies do not increase with compression as rapidly as they should.Figure15also shows the calculated temperature-dependence of the Gr¨u neisen parameter. At low temperatures(T<40K)the plot shows a fair amount of structure relative to the high-∼temperature curve.This can be understood from the phonon dispersion shown in Fig.5, where the lowest branch is in the[ξξξ]direction and becomesflat around3THz,frequencies that become relevant in their contribution to the specific heat at temperatures around a third of their energy,i.e.,around50K.This branch is the lowest and hence appearsfirst with increasing temperature,furthermore it appears with a lot of weight as there are eight spatial directions corresponding to these modes.At low temperatures the phonon contribution to the heat capacity is proportional to T3 and vanishes more rapidly than the electronic contribution,which is linear in temperature. Figure16shows the calculated coefficient of the electronic contribution to the heat capacity,π2γel=which is proportional to the density of states at the Fermi energy,n(E F).Compression of the crystal reduces n(E F),i.e.,γel decreases monotonically.IV.SUMMARYThe work presented here is aimed at(1)improving the tight-bindingfit of copper specif-ically for the calculation of thermodynamic properties,and(2)investigating the transfer-ability and range of applicability of the improved model.For the model to be reliable in calculating thermodynamic properties,it must produce a phonon dispersion in good agreement with experiment.The initial model wasfit to first-principles calculations of the total energy at a series of different volumes for the cubic crystal structures.The database offirst-principles calculations was extended here to include fcc structures distorted to reflect high-symmetry phonon modes and the elastic constants;fitting to the extended database yields the improved model which indeed delivers phonon frequencies significantly closer to the experimental values.From the phonon density of states the free energy was calculated,in the quasi-harmonic approximation,as a function of volume and temperature.The temperature-dependence of the minimum of the free energy directly yields the thermal expansion and the linear expansion coefficient,both in good agreement with experiment.The elastic constants are somewhat improved over the initial model,though discrepancies with experiment remain evident.The quantities in the previous paragraph depend on volumes only in the vicinity of the T=0equilibrium volume.The volumes used for the cubic and the distorted fcc structures in thefit extend over a wide range;the equilibrium volume is not treated any differently than other values(down to9.7˚A3,the smallest volume for which distorted structures were fit).This gives some confidence that the model applies to a range beyond the equilibrium volume and its immediate vicinity.Within the quasi-harmonic approximation the volume dependence of the phonon fre-quencies gives a non-zero Gr¨u neisen parameter;the results calculated from the TB model roughly agrees with the empiricalγ·ρ=constant.The magnitude is somewhat low,i.e., the compression-induced stiffening of the crystal remains somewhat weaker than is experi-mentally measured.The compression at which the model clearly fails can be seen from the Gr¨u neisen param-eter as well as the volume dependences of the elastic constants,the electronic contribution to the heat capacity,and the characteristic phonon temperatures.All of these entities vary monotonically with compression until the volume reaches approximately8˚A3,i.e.,a density of roughly13g/cm3,at which point unphysical behavior appears.The unphysical behavior points to the limitations of the model.The Hamiltonian and overlap matrix elements are described by a functional form which can at best approximate the actual behavior within a limited range.For an extended range either the functional form must be modified,e.g.by including higher-order terms in Eq.4,as has been done in a more recent NRL TB copper potential used in Ref.5.The need for modification can also be seen in the electronic band structure,which is degraded by thefitting to distorted fcc structures.V.ACKNOWLEDGMENTWe thank Jon Boettger,Matthias Graf,David Schiferl,and Duane Wallace for helpful and encouraging discussions.This research is supported by the Department of Energy un-der contract W-7405-ENG-36.All FLAPW calculations were performed using the Wien97 package.12Some of the calculations were performed at the National Energy Research Scien-tific Computing Center(NERSC),which is supported by the Office of Science of the U.S. Department of Energy under Contract No.DE-AC03-76SF00098REFERENCES1N.Troullier,J.L.Martins,Phys.Rev.B43,1993(1991).2C.Y.Wei,S.P.Lewis,E.J.Mele,and A.M.Rappe,Phys.Rev.B57,10062(1998).3F.Jona and P.M.Marcus,Phys.Rev.B63,094113/1(2001).4R.E.Cohen,M.J.Mehl,and D.A.Papaconstantopoulos,Phys.Rev.B50,14694(1994); M.J.Mehl and D.A.Papaconstantopoulos,Phys.Rev.B54,4519(1996);S.H.Yang, M.J.Mehl,and D.A.Papaconstantopoulos,Phys.Rev.B57,R2013(1998).5Y.Mishin,M.J.Mehl,D.A.Papaconstantopoulos,A.F.Voter,and J.D.Kress,Phys. Rev.B63,224106(2001).6W.J.Nellis,J.A.Moriarty,A.C.Mitchell,M.Ross,R.G.Dandrea,N.W.Ashcroft,N.C.Holmes,and G.R.Gathers,Phys.Rev.Lett.60,1414(1988).7H.K.Mao,P.M.Bell,J.W.Shaner,and D.J.Steinberg,J.Appl.Phys.49,3276(1978). 8R.G.McQueen,S.P.Marsh,J.W.Taylor,J.N.Fritz,and W.J.Carter,in High Velocity Impact Phenomena,R.Kinslow Ed.,(Academic,New York,1970).9S.G.Louie in Electronic Structure,Dynamics,and Quantum Structured Properties of Condensed Matter,edited by D.T.Devreese and P.van Camp(Plenum,New York, 1985),p.335.10J.C.Slater and G.F.Koster,Phys.Rev.94,1498(1954).11The(ll′m)symmetries are(ssσ),(spσ),(sdσ),(ppσ),(pdσ),(ddσ),(ppπ),(pdπ),(ddπ), and(ddδ),asfirst suggested by J.C.Slater and G.F.Koster,Phys.Rev.94,1498(1954). 12P.Blaha,K.Schwarz,and J.Luitz,WIEN97,A Full Potential Linearized Augmented Plane Wave Package for Calculating Crystal Properties,(Techn.Universit¨a t Wien,Aus-tria,1999).。

作者单位：清华大学建筑学院收稿日期：2013-08-01“城市翻修”教学系列报告（22）：意大利都灵手工卷烟厂改造城市设计Studio Teaching Reports on Urban Fixing (22): Urban Design of Manifattura Tabacchi Regeneration, Torino, Italy朱文一，万博，刘健/ZHU Wenyi, WAN Bo, LIU Jian“城市翻修”教学系列主持教师：朱文一 教授清华大学研究生建筑专业设计课程课程名称：建筑与城市设计（国际联合设计）课程编号：70000044课程时间：2013 年春季学期专题题目：意大利都灵手工卷烟厂改造城市设计合作学校：意大利都灵理工大学指导教师：清华大学：朱文一，张利，刘健，Michele Bonino ，万博（助教）都灵理工大学：Pierre-Alain Croset, Gustavo Ambrosini, Giacomo L. Beccaria （助教）清华大学选课学生：2012级研究生A 组：雷楠，孙昊德，袁晓宇B 组：姜文婷，王佳怡，冯思婕C 组：房宇巍，张丙生，杨睿D 组：崔敏，朱琳，范司琪都灵理工大学大学选课学生:A 组：Marco Carpani, Alessandro Desideri, Luca Gramaglia B 组：Arlind Dervishaj, Dario Palumeri, Alice Pinto, Michele Santarelli C 组：Erika Allegra, Giulia Demo, Qin Yun D 组：Gionatan Calvo, Kan Cheng, Mattia Grosso11 课程背景“意大利都灵手工卷烟厂改造城市设计”是清华大学与都灵理工大学两校建筑学院在2013年进行的联合设计课程教学（图1）。

清华大学建筑学院与都灵理工大学建筑学院的首次合作可以追溯到2008年，当时由清华大学建筑学院朱文一教授主持，两校开展了联合设计教学课程，并出版了反映该设计专题成果的书籍《北京奥运场馆赛后再利用研究》。

林格托工厂改建，都灵，意大利lingotto FactoRy conveRsion, toRino, italy, 1983-2003建筑设计：伦佐·皮亚诺建筑工作室ARCHITECTS: Renzo Piano Building Workshop4322 工厂原状：鸟瞰显示出建筑的纪念性/The factory as it was: aerial view shows the length and nobility of the machine become monument 3 历史照片：车辆在测试车道转弯处高速行驶/Historical image: cars speeding round corners of test track4 改造前：从汽车坡道望向屋顶/Original building prior to renovation: view up vehicular ramp to roof (2－4 © Lingotto)5 平面/Plan6 剖面/Section庭院4/Courtyard4庭院3/Courtyard3位于都灵的林格托工厂曾经是菲亚特汽车主要的生产场地，现在仍然是现代建筑最具代表性的工业建筑之一。

工程师马特·特鲁科从1915年开始了对工厂的设计，他从北美工业建筑中获得灵感，并沿用了亨尼比克曾在福特工厂的钢筋混凝土结构建造中使用的手法。

这座建筑位于尼斯米利方蒂区，坐落在尼斯路与都灵铁路的一条支路之间。

林格托工厂的体量达100万m3，长500m，高5层，是基于重复柱、梁、楼板这3种构成元素的模数化钢筋混凝土建筑的最早先例。

工作车间由两个超过500m长的、用于制造汽车的纵向元素构成，它们之间又通过5个容纳员工设施的多层横向元素连接。

1924年－1926年间，在纵向元素的端头增加了两个螺旋坡道。

地面层的汽车可以借此坡道直达建筑屋顶上的测试车道。

外國工程師傳授的design資料，供大伙儿享受。

Injection Molding Design Guidelines注塑模具设计指导Much has been written regarding design guidelines for injection molding. Yet, the design guidelines can be summed up in just a few design rules.已经有许多关于设计注塑模具的书了,可是设计准那么能够被归纳以下几点:Use uniform wall thicknesses throughout the part. This will minimize sinking, warping, residual stresses, and improve mold fill and cycle times.1.零件整体壁厚维持均匀.如此能够最小化缩坑,翘曲,强度减小,模具填充和循环时刻.Use generous radius at all corners. The inside corner radius should be a minimum of one material thickness.2.在转角处大量采纳圆角.圆角内部最小半径为一个壁厚.the least thickness compliant with the process, material, or product design requirements. Using the least wall thickness for the process ensures rapid cooling, short cycle times, and minimum shot weight. All these result in the least possible part cost.3.依照工艺,材料,产品设计需求.采纳最小壁厚.最小壁厚的应用能够确保外國工程師傳授的design資料，供大伙儿享受。

101 Innovation Drive San Jose, CA 95134(408) Designing with Low-Level PrimitivesUser GuideSoftware Version 7.1Document Version: 3.0 Document Date:April 2007Copyright © 2007 Altera Corporation. All rights reserved. Altera, The Programmable Solutions Company, the stylized Altera logo, specific device des-ignations, and all other words and logos that are identified as trademarks and/or service marks are, unless noted otherwise, the trademarks and service marks of Altera Corporation in the U.S. and other countries. All other product or service names are the property of their respective holders. Al-tera products are protected under numerous U.S. and foreign patents and pending applications, maskwork rights, and copyrights. Altera warrants performance of its semiconductor products to current specifications in accordance with Altera's standard warranty, but reserves the right to make changes to any products and services at any time without notice. Altera assumes no responsibility or liability arising out of the ap-plication or use of any information, product, or service described herein except as expressly agreed to in writing by AlteraCorporation. Altera customers are advised to obtain the latest version of device specifications before relying on any published in-formation and before placing orders for products or services.UG-83105-3.0ContentsAbout this User Guide (v)How to Contact Altera (v)Typographic Conventions (vi)Chapter 1. Low-Level Primitive DesignIntroduction............................................................................................................................................1–1 Low-Level Primitive Examples...........................................................................................................1–2 LCELL Primitive...............................................................................................................................1–2 Using I/Os.........................................................................................................................................1–6 Using Registers in Altera FPGAs...................................................................................................1–7 Creating Memory for Your Design................................................................................................1–9 Look-Up Table Buffer Primitives.................................................................................................1–13 Chapter 2. Primitive ReferencePrimitives................................................................................................................................................2–1 ALT_INBUF......................................................................................................................................2–1 ALT_OUTBUF..................................................................................................................................2–3 ALT_OUTBUF_TRI..........................................................................................................................2–6 ALT_IOBUF.......................................................................................................................................2–8 ALT_INBUF_DIFF.........................................................................................................................2–11 ALT_OUTBUF_DIFF.....................................................................................................................2–13 ALT_OUTBUF_TRI_DIFF.............................................................................................................2–14 ALT_IOBUF_DIFF..........................................................................................................................2–19 ALT_BIDIR_DIFF...........................................................................................................................2–22 ALT_BIDIR_BUF............................................................................................................................2–25 LCELL..............................................................................................................................................2–27 DFF...................................................................................................................................................2–28 CARRY and CARRY_SUM...........................................................................................................2–29 CASCADE.......................................................................................................................................2–30 LUT_INPUT....................................................................................................................................2–31 LUT_OUTPUT................................................................................................................................2–32 Synthesis Attributes............................................................................................................................2–33ContentsAbout this User Guide DocumentRevision HistoryThe table below shows the revision history for this document.How to ContactAlteraFor the most up-to-date information about Altera® products, go to theAltera world wide web site at . For technical support onthis product, go to /mysupport. For additionalinformation about Altera products, consult the sources shown below.Date andDocumentVersionChanges Made Summary of ChangesApril 2007v3.0Made changes to the Guide:●Added examples 1–8, 2–4, 2–6, 2–8●Added these new sections:“ALT_INBUF_DIFF” on page2–11“ALT_OUTBUF_DIFF” on page2–13“ALT_OUTBUF_TRI_DIFF” on page2–14“ALT_IOBUF_DIFF” on page2–19“ALT_BIDIR_DIFF” on page2–22“ALT_BIDIR_BUF” on page2–25●Removed most of the “Synthesis Attributes” on page2–33section, replaced with a reference to the Quartus IIHandbook.Technical changes tocoincide with changes tothe Quartus II software7.0.0 releaseMay 2006v2.0Technical changes to coincide with changes to the Quartus IIsoftware 6.0.0 release—October 2005v1.0Initial Release—Information Type ResourceT echnical support /mysupport/Product literature Altera literature services literature@ (1)FTP site Note to table:(1)You can also contact your local Altera sales office or sales representative.Typographic ConventionsTypographicConventionsThis document uses the typographic conventions shown below.Visual Cue MeaningBold Type with Initial Capital Letters Command names, dialog box titles, checkbox options, and dialog box options are shown in bold, initial capital letters. Example: Save As dialog box.bold type External timing parameters, directory names, project names, disk drive names,filenames, filename extensions, and software utility names are shown in boldtype. Examples: f MAX, \qdesigns directory, d: drive, chiptrip.gdf file.Italic Type with Initial Capital Letters Document titles are shown in italic type with initial capital letters. Example: AN 75: High-Speed Board Design.Italic type Internal timing parameters and variables are shown in italic type.Examples: t PIA, n + 1.Variable names are enclosed in angle brackets (< >) and shown in italic type.Example: <file name>, <project name>.pof file.Initial Capital Letters Keyboard keys and menu names are shown with initial capital letters. Examples:Delete key, the Options menu.“Subheading Title”References to sections within a document and titles of on-line help topics areshown in quotation marks. Example: “T ypographic Conventions.”Courier type Signal and port names are shown in lowercase Courier type. Examples: data1,tdi, input. Active-low signals are denoted by suffix n, e.g., resetn.Anything that must be typed exactly as it appears is shown in Courier type. Forexample: c:\qdesigns\tutorial\chiptrip.gdf. Also, sections of anactual file, such as a Report File, references to parts of files (e.g., the AHDLkeyword SUBDESIGN), as well as logic function names (e.g., TRI) are shown inCourier.1.,2.,3., anda.,b.,c., etc.Numbered steps are used in a list of items when the sequence of the items is important, such as the steps listed in a procedure.■●•Bullets are used in a list of items when the sequence of the items is not important. v The checkmark indicates a procedure that consists of one step only.1The hand points to information that requires special attention.c A caution calls attention to a condition or possible situation that can damage ordestroy the product or the user’s work.w A warning calls attention to a condition or possible situation that can cause injuryto the user.r The angled arrow indicates you should press the Enter key.f The feet direct you to more information on a particular topic.1.Low-Level PrimitiveDesignIntroduction Your hardware description language (HDL) coding style can have asignificant effect on the quality of results that you achieve forprogrammable logic designs. Although synthesis tools optimize HDLcode for both logic utilization and performance, sometimes the bestoptimizations require engineering knowledge of the design. Therefore, itis important to consider the HDL coding style that you adopt whencreating your programmable logic design.Low-level HDL design is the practice of using low-level primitives andassignments in your HDL code to dictate a particular hardwareimplementation for a piece of logic. Low-level primitives are smallarchitectural building blocks that assist you in creating your design. Withthe Quartus®II software, you have the option of using low-level HDLdesign techniques that can help you to achieve better resource utilizationor faster timing results.Using low-level primitives in your design enables you to control thehardware implementation for a cone of logic in your design. These conescan be as small as an LCELL instantiation, which prevents the Quartus IIsynthesis engine from performing optimizations, to larger, more complexexamples that specify the encoding method for a finite state machine(FSM).The Quartus II software can synthesize and place and route designs thatinstantiate low-level primitives. This user guide describes the supportthat the Quartus II software offers for creating a design with primitivesand includes the definition of each primitive, usage guidelines, andexample designs.Using the Quartus II software, you can instantiate a Quartus II primitiveinto your HDL design. The source files for Quartus II primitives are builtinto the Quartus II software.Low-Level Primitive ExamplesExample1–1 is a small Verilog example that shows an instantiation of aDFF primitive and an ALT_OUTBUF_TRI primitive.Example1–1.Instantiation of a DFF Primitive and alt_outbuf_tri Primitive, Verilogmodule compinst (data, clock, clearn, presetn,a, b, q_out, t_out);input data, clock, clearn, presetn, a, b;output q_out, t_out;dff dff_inst (.d (data), .q (q_out), .clk (clock),//dff is a primitive.clrn(clearn),.prn (presetn));alt_outbuf_tri tri_inst (.i(b), .oe(a), .o(t_out))// alt_outbuf_tri is a primitiveendmoduleLow-Level Primitive Examples The following sections provide examples of how you can implement low-level primitives:■“LCELL Primitive”■“Using I/Os” on page1–6■“Using Registers in Altera FPGAs” on page1–7■“Creating Memory for Your Design” on page1–9■“Look-Up Table Buffer Primitives” on page1–13For detailed specification of the primitive’s ports used in these sections, refer to “Primitives” on page2–1.LCELL PrimitiveThe LCELL primitive allows you to break up your design into manageable parts and prevents the Quartus II synthesis engine from merging logic. This is especially useful when you are trying to debug your design at the implementation level.Low-Level Primitive DesignIn Example1–2, the LCELL primitive separates the logic in your design.The first code example and the resulting view from the Quartus IITechnology Map Viewer (Figure1–1) show that the logic is mergedduring the synthesis process.Example1–2.LCELL Primitive Separates Logicmodule logic_merge(clk,addr,data,dataout);input clk;input [3:0] addr;input [2:0] data;output[2:0] dataout;reg[2:0] dataout;wire temp_0;wire temp_1;wire temp_2;wire temp_3;wire temp_4;wire temp_5;wire temp_6;assign temp_3 = addr[0] & addr[1] & addr[2] & addr[3];assign temp_4 = addr[3] & addr[2] & addr[1] & addr[0];assign temp_1 = addr[1] & addr[2] & addr[3];assign temp_2 = temp_1 & addr[0];assign temp_5 = temp_2 & data[0];assign temp_6 = temp_3 & data[1];assign temp_0 = temp_4 & data[2];always@(posedge clk)begindataout[2] <= temp_0;endalways@(posedge clk)begindataout[0] <= temp_5;endalways@(posedge clk)begindataout[1] <= temp_6;endendmoduleLow-Level Primitive ExamplesFigure1–1.Logic Merged During the ProcessBy strategically placing the LCELLs, you can control how the Quartus IIsynthesis engine splits your design into logic cells. This typically causesyour design to use more logic resources, so this primitive should be usedwith care. In the following example, and the resulting view from theQuartus II Technology Map Viewer (Figure1–2), three LCELL primitiveinstantiations are introduced between the combinational logic. Note that“LCELL” is also the name that the Technology Map Viewer gives to thelogic cells in some device families, as shown in the figure.In Example1–3, the address decoder logic is not merged with theregisters in the design.Example1–3.Address Decoder Logic Not Merged with Registersmodule comb_logic_with_lcells(clk,addr,data,dataout);input clk;input[3:0] addr;input [2:0] data;output [2:0] dataout;reg[2:0] dataout;wire temp_0;wire temp_1;wire temp_2;wire temp_3;wire temp_4;wire temp_5;wire temp_6;wire temp_7;wire temp_8;wire temp_9;assign temp_1 = addr[0] & addr[1] & addr[2] & addr[3];assign temp_3 = temp_4 & addr[0];assign temp_8 = temp_5 & data[0];assign temp_9 = temp_6 & data[1];assign temp_0 = temp_7 & data[2];assign temp_4 = addr[1] & addr[2] & addr[3];assign temp_2 = addr[3] & addr[2] & addr[1] & addr[0];lcell inst1(.in(temp_1),.out(temp_6));lcell inst2(.in(temp_2),.out(temp_7));lcell inst3(.in(temp_3),.out(temp_5));always@(posedge clk)begindataout[2] <= temp_0;endalways@(posedge clk)begindataout[0] <= temp_8;endalways@(posedge clk)begindataout[1] <= temp_9;endendmoduleFigure1–2.LCELL Primitive InstantiationsUsing I/OsWith I/O primitives, you can make I/O assignments in your HDL fileinstead of making them through the Assignment Editor in the Quartus IIsoftware. Example1–4 describes how to make an I/O standardassignment to an input pin using the ALT_INBUF primitive in VerilogHDL.Example1–4.Making an I/O Standard Assignment to an Input Pin Using the ALT_INBUF Primitive, Verilog module io_primitives (data_in, data_out);input data_in;wire internal_sig;output data_out;alt_inbuf my_inbuf (.i(data_in), .o(internal_sig));defparam my_inbuf.io_standard="1.8V HSTL Class I";assign data_out = !internal_sig;endmoduleFor detailed specifications of the primitive’s ports used in these sections,refer to “Primitives” on page2–1.I/O AttributesThere are no primitives available to define an I/O register that can beimplemented as a fast input, fast output, or fast output enable register.However, registers associated with an input or output pin can be movedinto I/O registers using the following assignments in the Quartus IIsoftware for those I/O pins:■fast_input_register■fast_output_register■fast_output_enable_registerThese assignments can be set by HDL synthesis attributes. Example1–5illustrates the fast_output_register synthesis attribute.Example1–5.The fast_output_register Synthesis Attributemodule fast_output(i,clk,o);input i;output o;reg o /* synthesis altera_attribute = ”FAST_OUTPUT_REGISTER”=ON */;always @(posedge clk)begino <= i;endendmodule1For more information, refer to “Synthesis Attributes” onpage2–33.Using Registers in Altera FPGAsThe building blocks of FPGA architectures contain a combinationalcomponent along with a register component. Each register component inan Altera FPGA provides a number of secondary control signals (such asclear, reset, and enable signals) that you can use to implementcontrol logic for each register without the use of extra logic cells. Devicefamilies vary in their support for secondary signals, so you must consultthe device family data sheet to verify which signals are available in yourtarget device. Download the device family data sheets from the Literaturesection of .Inferring Registers Using HDL CodeTo make the most efficient use of the signals in the device, your HDL code should match the device architecture as closely as possible. Because of the layout of the architecture, the control signals have a certain priority. Therefore, your HDL code should follow that priority whenever possible. If you do not follow the signal priority, your synthesis tool emulates the control signals using additional logic resources. Therefore, creating functionally correct results is always possible. However, if your design requirements are flexible (in terms of which control signals are used and in what priority), you can match your design to the target device architecture to achieve the optimal performance and logic utilization results.There are certain cases where using extra logic resources to emulate control signals can have an unintended impact. For example, aclock_enable signal has priority over a synchronous_reset or a clear signal in the device architecture. The clock_enable signal disables the clock line in the logic array block (LAB), and the sclr signal is synchronous. In the device architecture, the synchronous clear takes effect only when a clock edge occurs.If you code a register with a synchronous clear signal that has priority over a clock enable signal, the software must emulate the clock enable functionality using data inputs to the registers. Because the signal does not use the clock enable port of a register, you cannot apply a Clock Enable Multicycle constraint. In this case, following the priority of signals available in the device is clearly the best choice for the priority of these control signals because using a different priority causes unexpected results with an assignment to the clock enable signal. The signal order is the same for all Altera device families, although, as mentioned earlier, not all device families provide every signal. In general, use the signal order shown in Table1–1.Table1–1.Signal Order (from Highest to Lowest Priority)Priority Signal1Asynchronous clear2Preset3Asynchronous load4Enable5Synchronous clear6Synchronous load7Data inThe sclr signal is not inferred by Quartus Integrated Synthesis whenthere are a large number of registers with different sclr signals.Thisbehavior makes it easier for the fitter to successfully route the design. Ifyou would like to force the use of the sclr signals, you can use thefollowing Quartus II synthesis settings.f For more details about these and other synthesis settings, refer to theQuartus II Integrated Synthesis chapter in volume1 of the Quartus IIHandbook.■Force Use of Synchronous Clear Signals—Forces the compiler to utilize synchronous clear signals in normal mode logic cells.Turning on this option helps to reduce the total number of logic cellsused in the design, but might negatively impact the fitting becausesynchronous control signals are shared by all the logic cells in a LAB.■Allow Synchronous Control Signals—Allows the compiler to utilize synchronous clear and/or synchronous load signals in normalmode logic cells. Turning on this option helps to reduce the totalnumber of logic cells used in the design, but might negatively impactthe fitting because synchronous control signals are shared by all thelogic cells in a LAB.f For more information about inference guidelines for registers and onsecondary control signal inference rules, refer to the Recommended HDLCoding Styles chapter in volume1 of the Quartus II Handbook.Using the DFFEAS PrimitiveThe DFFEAS primitive allows you to directly instantiate a register in yourdesign and gives you control over which secondary signals are used. TheDFFEAS primitive instantiations are always adhered to unless thesecondary control signals that you use are not supported by the devicefamily architecture. If you instantiate a DFFEAS primitive withunsupported secondary control signals, they are converted into theequivalent logic.1For an example on instantiation of the DFFEAS primitive, refer to the Primitive Reference and Synthesis Attributes chapter in thisuser guide.Creating Memory for Your DesignYou can create RAM for your design in two ways. The first methodinvolves creating HDL code that infers a memory function. The secondmethod involves building a function using the MegaWizard® Plug-InManager and instantiating the resulting custom megafunction variationfile in your design.Inferring RAM Functions from HDL CodeTo infer RAM functions, synthesis tools detect sets of registers and logicthat can be replaced with the altsyncram or lpm_ram_dp megafunctions,depending on the targeted device family. The Quartus II software usuallydoes not infer very small RAM blocks because they typically areimplemented more efficiently by using the registers in regular logic.If your design contains a RAM block that your synthesis tool does notrecognize and infer, the design might require a large amount of systemmemory, which can potentially cause run-time compilation problems.f For RAM inference guidelines, refer to the Recommended HDL CodingStyles chapter of the Quartus II Handbook.Using the MegaWizard Plug-In ManagerYou can use the MegaWizard Plug-In Manager to create RAM functions.The MegaWizard Plug-In Manager, located in the Tools menu in theQuartus II software, allows you create or modify design files that containcustom megafunction variations, which you can then instantiate in adesign file.The GUI-based interface of the MegaWizard Plug-In Manager providesan easy and intuitive interface that allows you to parameterize complexfunctions such as memory. However, there are cases, particularly withmemory, where you simply want to modify a small component of themegafunction. For example, your design can call for two types of memoryfunctions: a 32, 8-bit word single-port memory function and a 64, 8-bitword single-port memory function. In this scenario, you can use theMegaWizard Plug-In Manager to create one function and then use theinstantiation from the wizard-generated file to directly instantiate thesecond variation. However, directly instantiating memory functionsshould only be used when the modifications to the functions are minimal.Example1–6 shows a Verilog example for a 32, 8-bit word single-portmemory function.Example1–6.A 32, 8-Bit Word Single-Port Memory Function, Verilogaltsyncramalt syncram_component (.wren_a (wren),.clock0 (clock),.address_a (wraddress),.address_b (rdaddress),.data_a (data_in),.q_b (data_out),.aclr0 (1'b0),.aclr1 (1'b0),.clocken1 (1'b1),.clocken0 (1'b1),.q_a (),.data_b ({8{1'b1}}),.rden_b (1'b1),.wren_b (1'b0),.byteena_b (1'b1),.addressstall_a (1'b0),.byteena_a (1'b1),.addressstall_b (1'b0),.clock1 (1'b1));defparamaltsyncram_component.address_aclr_a = "NONE",altsyncram_component.address_aclr_b = "NONE",altsyncram_component.address_reg_b = "CLOCK0",altsyncram_component.indata_aclr_a = "NONE",altsyncram_component.intended_device_family = "Stratix",altsyncram_component.lpm_type = "altsyncram",//This is where a 32, 8-bit word is modified.altsyncram_component.numwords_a = 32,altsyncram_component.numwords_b = 32,altsyncram_component.operation_mode = "DUAL_PORT",altsyncram_component.outdata_aclr_b = "NONE",altsyncram_component.outdata_reg_b = "CLOCK0",altsyncram_component.power_up_uninitialized = "FALSE",altsyncram_component.read_during_write_mode_mixed_ports = "DONT_CARE",altsyncram_component.widthad_a = 5,altsyncram_component.widthad_b = 5,//This is the width of the input port.altsyncram_component.width_a = 8,altsyncram_component.width_b = 8,altsyncram_component.width_byteena_a = 1,altsyncram_component.wrcontrol_aclr_a = "NONE";Example1–7 shows a Verilog example for a 64, 8-bit word single-portmemory function.Example1–7.A 64, 8-Bit Word Single-Port Memory Function, Verilogaltsyncram altsyncram_component (.wren_a (wren),.clock0 (clock),.address_a (wraddress),.address_b (rdaddress),.data_a (data_in),.q_b (data_out),.aclr0 (1'b0),.aclr1 (1'b0),.clocken1 (1'b1),.clocken0 (1'b1),.q_a (),.data_b ({8{1'b1}}),.rden_b (1'b1),.wren_b (1'b0),.byteena_b (1'b1),.addressstall_a (1'b0),.byteena_a (1'b1),.addressstall_b (1'b0),.clock1 (1'b1));defparamaltsyncram_component.address_aclr_a = "NONE",altsyncram_component.address_aclr_b = "NONE",altsyncram_component.address_reg_b = "CLOCK0",altsyncram_component.indata_aclr_a = "NONE",altsyncram_component.intended_device_family = "Stratix",altsyncram_component.lpm_type = "altsyncram",//This is where a 64, 8-bit word is modified.altsyncram_component.numwords_a = 64,altsyncram_component.numwords_b = 64,altsyncram_component.operation_mode = "DUAL_PORT",altsyncram_component.outdata_aclr_b = "NONE",altsyncram_component.outdata_reg_b = "CLOCK0",altsyncram_component.power_up_uninitialized = "FALSE",altsyncram_component.read_during_write_mode_mixed_ports="DONT_CARE",altsyncram_component.widthad_a = 6,altsyncram_component.widthad_b = 6,//This is the width of the input port.altsyncram_component.width_a = 8,altsyncram_component.width_b = 8,altsyncram_component.width_byteena_a = 1,altsyncram_component.wrcontrol_aclr_a = "NONE";Look-Up Table Buffer PrimitivesThe look-up table (LUT) buffer primitives, LUT_INPUT andLUT_OUTPUT, specify a LUT function in your design. These primitivesare single-input, single-output buffers that you use to create a LUTdirectly in your design. The LUT_INPUT acts as an input to theLUT_OUTPUT. If your design contains a LUT_OUTPUT primitive that isnot properly driven, the LUT_OUTPUT is ignored. By using LUT_INPUTand LUT_OUTPUT, you can specify which LUT inputs are used. Theseprimitives are similar to the LCELL primitive; they give you control overhow the Quartus II synthesis engine breaks your design up into logiccells. Because they give you full control of the inputs and outputs to alogic cell, LUT_INPUT and LUT_OUTPUT primitives give you morecontrol over the synthesis process, but you must have moreunderstanding of the device architecture to use them successfullyExample1–8 shows a primitive instantiation that creates a four-inputLUT that implements the function aw & bw ^ cw | dw.Example1–8.A Primitive Instantiation that Creates a Four-Input LUTmodule lut_function (a,b,c,d,o);input a,b,c,d;output o;wire aw,bw,cw,dw,o;lut_input lut_in1 (a, aw) ;lut_input lut_in2 (b, bw) ;lut_input lut_in3 (c, cw) ;lut_input lut_in4 (d, dw) ;lut_output lut_o (aw & bw ^ cw | dw, o) ;endmodule。

Published in Proceedings of the Seventeenth Annual Cognitive Science Conference,Pittsburg,PA,July1995(pp.78–83)Opportunistic Reasoning:A Design PerspectiveMarin D.SiminaCollege of Computing Georgia Institute of Technology Atlanta,GA30332-0280 marin@Janet L.KolodnerCollege of Computing Georgia Institute of Technology Atlanta,GA30332-0280jlk@AbstractAn essential component of opportunistic behavior is oppor-tunity recognition,the recognition of those conditions that facilitate the pursuit of some suspended goal.Opportunity recognition is a special case of situation assessment,the pro-cess of sizing up a novel situation.The ability to recognize opportunities for reinstating suspended problem contexts(one way in which goals manifest themselves in design)is crucial to creative design.In order to deal with real world oppor-tunity recognition,we attribute limited inferential power to relevant suspended goals.We propose that goals suspended in the working memory monitor the internal(hidden)represen-tations of the currently recognized objects.A suspended goal is satisfied when the current internal representation and a sus-pended goal“match”.We propose a computational model for working memory and we compare it with other relevant theo-ries of opportunistic planning.This working memory model is implemented as part of our IMPROVISER system.IntroductionDuring a mechanical engineering project a group of students were asked to design and implement a mechanical device for the quick and safe transportation of a fragile cargo(some eggs).The students went to a Home Depot(a hardware store), where they started by choosing springs for the launching com-ponent.During the design process they made the following observations:Andy:hey,when I compress the spring it bends; this weakens the force of the springsMary(wrapping her hand around the spring):yes, we have to enclose it in a tubeBill:the tube should be collapsible,otherwise the spring cannot be compressedThe students began proposing mechanisms thatfit this de-scription.One of them suggested a telescope,but it was rejected by the group because it was expected to be costly and it did notfit in the available budget.Another student proposed a collapsible camping cup,which unfortunately has a wrong shape.The designers were unable to think of where in the store they might look for a useful collapsible tube,so they moved to another part of their problem.They started thinking about load protection.Since sponges are a good way to pro-vide cushioning,they decided to go to the store’s bathroom section.During the search for sponges,one of the students saw a toilet paper holder and exclaimed:Mary:Look!A collapsible tube!The whole group agreed that the toilet paper holder fulfilled the requirements of their previously suspended problem. The above example illustrates a rather mundane,but com-mon,experience in doing design.The students started by structuring the initial problem(launching,cushioning)and then they tried to elaborate the subcomponents,one at a time. When they were stuck with one subproblem,they suspended it,and they approached another related subproblem.When they saw the toilet paper holder,however,they recognized that an opportunity to address the suspended subproblem had presented itself in the environment.What processes are responsible for recognizing such oppor-tunities?How can a cognitive architecture handle this kind of processing?What constraints are there,if any,on the work-ings of these processes?We are studying these problems in the context of developing a cognitive model for creative design. Our computer program,IMPROVISER(Wills&Kolodner 1994b),was extended in order to help us answer the above questions.Our exploration of creative design(Kolodner&Wills 1993a)suggests that the conceptual phase,in which the prob-lem is framed,plays a key role in designing.In this phase, which is interspersed throughout the design process,the prob-lem situation is assessed and the given problem is reformu-lated and restructured.While one can organize the subgoals involved in conceptual design in a hierarchical structure,the processing of these subgoals seems far more unstructured. Designers often begin by proposing a shallow hierarchical set of subgoals as they initially formulate the way they will solve a problem(e.g.,the artifact we are designing has these n parts or mechanisms;we need to design each one).They continue by addressing each of the subgoals,one at a time. It is here where the organized reasoning breaks down.When the designer fails to solve one subproblem,he/she seems to suspend it and approach another related subproblem(as in the example above).Sometimes the next subproblem is simply a not-yet-considered sibling subgoal(as when the student de-signers moved from designing their spring launch mechanism to the cushioning for the eggs);sometimes the opportunity to go back to a suspended subgoal is recognized(as when the toilet paper holder was seen).When we consider the incremental and recursive nature of this reasoning process,we can identify one way of recogniz-ing that a previously-suspended subgoal might be successfully addressed.During consideration of a new subproblem,the de-signer has to consider interactions with related subproblems, some of which have been suspended previously.This mayprovide a fresh view of the suspended problem and a new way to redescribe it.Redescription or new insights about a sub-problem gained during reasoning trigger the goal scheduler to unblock the suspended subproblem,allowing already-known solutions to be recalled or new means of solving it to be rec-ognized.This means of unblocking a suspended goal is com-pletely under the control of the reasoner,which knows which subproblems have been part of its most recent reasoning. But recognizing in the toilet paper holder the opportunity to address a suspended goal requires additional mechanisms that scan the environment and recognize when the environment is providing new insights into suspended goals.If the number of suspended goals,the complexity of the environment,or the amount of newness in the environment is high,such a mechanism could easily be overwhelmed.The mechanisms that provide this capability must be able to deal with such complexity.Opportunity RecognitionThe ProblemThe prerequisite for opportunistic behavior is the existence of suspended goals(problems),goals that cannot be pursued in the current context and are postponed.An essential component of opportunistic behavior is oppor-tunity recognition,recognition of those conditions that facil-itate the pursuit of some suspended goal.But opportunities seem to appear when they are not expected.The student de-signers,for example,had not previously thought about a toilet paper holder functioning as a collapsible tube.Recognizing the opportunitymeant both noticing the toilet paper holder and recognizing that its mechanism(which is hidden)included a collapsible tube.More than a simple matching mechanism is needed.Birnbaum(1986)suggests two central problems that must be addressed by a theory of opportunistic behavior:(1)how to detect opportunities and(2)how to“activate”the goals to which they pertain.An important issue here is identifying how much and what kind of processing is required in order to recognize the presence of the features that constitute an opportunity.A Critical ReviewHayes-Roth&Hayes-Roth(1979)proposed thefirst signifi-cant cognitive model of opportunistic behavior.Their model of opportunistic planning was inspired by protocols of sub-jects planning a hypothetical day’s errands.But they were most concerned with planning methods and gave little atten-tion to recognition processes.In fact,the experimenter always mentioned opportunitiesto the subjects when they overlooked them,and the subjects never tried their plans in the real world, so they never really dealt with genuine opportunities and the problem of recognizing them.Birnbaum(1986)gave more attention to recognition issues. He proposed the mental notes model,in which whenever a goal cannot be immediately satisfied,it is indexed in terms of the unmet preconditions that prevented its satisfaction.How-ever,as he points out,if the goal is indexed too specifically, then there will be many cases in which it will not be recalled even though an opportunity for its satisfaction is present,and if the goal is indexed in terms of more abstract features,we cannot assume that the agent will automatically generate the abstract description that will activate the goal.In order to solve the above dilemma within the framework of the mental notes model,Birnbaum suggests1spending some effort,when the goal is formed,to determine the range of situations in which it might easily be satisfied–for example, by constructing several incomplete plans for the goal in order to identify the relevant preconditions–and then indexing the goal in terms of the features that might arise in such situa-tions.Birnbaum&Collins(1984)also suggest an active goal framework,where all the goals have the ability to examine the current situation and to initiate inference to test their own relevance.Patalano,Seifert and Hammond(1992)criticize the use of active goals proposed by Birnbaum&Collins,claiming that this approach to opportunistic behavior is an unlikely expla-nation of human cognitive processes because of its computa-tional demands.However,Patalano,Seifert and Hammond do pick up on Birnbaum’s indexing scheme,calling it predictive encoding.Predictive encoding stresses the importance of en-coding blocked goals in memory in such a way that they will be recalled by conditions favorable for their solutions.Their experimental results show evidence of this process. However,the predictive encoding hypothesis seems incom-plete,because it does not enable a cognitive agent to recognize opportunities other than those which it is able to anticipate. In particular,it does not enable an agent to recognize novel opportunities,which by their very nature,cannot be easily anticipated.Recognition of the toilet paper holder as a col-lapsible tube,for example,is novel in that this is not the way a toilet paper holder is generally thought of.Similar issues caused Birnbaum&Collins(1984)to conclude that if an op-portunity is to be detected at all,inferential resources must be allocated to the goal recognition task.Ram&Hunter(1992)suggested a balance between back-ward chaining at the time of goal suspension and forward chaining at the time of opportunity recognition.In AQUA,a set of utility metrics have been proposed in order to make a tradeoff between predictive encoding and active goals.Un-fortunately,these utility metrics are very specific to story understanding.This suggests that we need active goals in order to recog-nize novel opportunities,but we need to control their power and number to make them computationally feasible.We need predictive encoding,but we also need more powerful infer-ential capabilities.We hope that an analysis of the exam-ple presented previously can provide insight in formulating a mechanism with these properties.A Possible SolutionWhy did the students fail to remember the toilet paper holder when they were trying to decide where they mightfind a collapsible tube,and what allowed them to recognize it as appropriate when they saw it?One possible reason why the toilet paper holder was not re-called and considered while thinking about collapsible tubes is that the probe that had been constructed(i.e.,the item de-scription used for remembering)was incompletely specified. Consequently,they retrieved items that fulfilled primary butnot secondary characteristics of the probe (e.g.,a telescope costs too much and a camping cup has a wrong shape).Af-ter every retrieval and evaluation of a new device,the probe was respecified,taking into consideration the initially ignored constraints (e.g.,we want something like a telescope,but cheaper).This process was suspended,however,before the toilet paper holder was recalled.But why was the probe inadequate for retrieving such a common object as a toilet paper holder?Our explanation is that the toilet paper holder is routinely associated with what its purpose is in the bathroom (holding toilet paper rolls)rather than with how this function is achieved (by means of a collapsible tube with a spring inside).It is not a particularly interesting device,and even though we see it every day,most of it is hidden by the roll of paper.Research shows that it is quite difficult to overcome such functional fixedness (Mayer 1970),which associates everyday objects with their obvious function (holding a paper roll in the case of the toilet paper holder).Routinely,we ignore other potential uses that can be derived from the structure and behavior of such everyday objects.Once we have specified desired criteria in a probe,it is easy to check them against a specific object.But if those criteria are different than those used to describe an object in memory,recall won’t happen.C1C2Behavioral Properties:Structural Properties:Perceivable Properties:Functional Properties:Use: Hold Paper Roll Probe Description: Parts: Cylinders C1 and C2; Spring S Fits−Inside(C1, C2) C1: Solidity(Hollow) C2: Solidity(Hollow)Composition of Cylinders (C1C2) Solidity(Hollow)Length = Length(S) + delta Shape = Cylindrical Enclosed(S, C1C2)Case Content:Case Index:States: Steady, Squeezed, Rest Steady:Length(S) < Rest−Length(S)Length(C1C2) = Width(Wall−Fixture)Squeezed:Length(C1C2) < Width(Wall−Fixture) Length(C1C2) < Length(C2)Rest:Length(C1C2) > Width(Wall−Fixture) Length(S) = Rest−Length(S)Rigid−Tube:Shape = Cylindrical Length−Variability = variesRadius−Variability = constant ...Figure 1:The many representations of a toilet paper holder Figure 1shows this mismatch.The collapsible tube,asdescribed after manipulating the springs (see the P ROBE D E -SCRIPTION in Figure 1),has the structural property that its shape is cylindrical and the behavioral property that its length can vary .The toilet paper holder,on the other hand,is indexed in memory by a combination of its Functional Properties and Perceivable Properties ,shown as I NDEX in Figure 1.Thus,we cannot retrieve the C ASE C ONTENT ,namely the Structural Properties and Behavioral Properties of the toilet paper holder by using the P ROBE D ESCRIPTION .What facilitates recognition of the opportunity in the envi-ronment,i.e.,recognition that the toilet paper holder can fulfill the role of collapsible tube?On the store’s shelf,one can see the shape of the device.Recognition procedures perceive that it is a collapsible tube,which matches the description from the retrieval probe and presumably the label that designates what needs to be encountered to unblock the suspended goal.But what processes direct recognition procedures to attend to the toilet paper holder on the store’s shelf?And what mechanisms allow matching of something in the environment to a goal that is no longer active?We know that memory search is incremental and that when our memories can’t re-trieve what we are asking them for,we redescribe what we are looking for and try again.But when we aren’t making head-way,we postpone additional retrieval until more information is gathered and pursue other retrieval strategies or subgoals (Williams &Hollan,1981,Norman &Bobrow,1979,Kolod-ner,1984).Similarly (and implied by predictive encoding),we suspend reasoning subgoals and subproblems that depend on postponed retrieval strategies and unmatched probes,mark-ing them with criteria that,if encountered,predict that they should be reopened (Patalano,Seifert and Hammond 1992).We propose that when an active subgoal (subproblem)is suspended,the subgoal and its criteria remain in working memory’s working store for some limited time.We further propose that goals suspended in the working memory con-tinuously monitor the environment,looking for matches in the environment to the specified criteria.Furthermore,we suggest that there are only a small number of these active goals.A computational model will provide more detail on these limitations.A Memory ModelThe Memory ArchitectureThe major component of our computational model (presented in Figure 2)is a working memory (WM),which communi-cates with both long-time memory (LTM)and perceptual pro-cesses and keeps track of recent reasoning context.As Barsa-lou (1992)suggests,the working memory mediates between short-term memory (STM)and the activated part of LTM.But we add significantly to Barsalou’s conception.First,we give the WM a structure.Second,the structure integrates components of STM with activated portions of LTM and with perceptual mechanisms and stores.Third,this integrated com-ponent acts as a buffer for LTM.It is the place where LTM’s components are manipulated and adapted.Fourth,we add a control unit (matcher),which can match (1)the current arti-fact being reasoning about or (2)all the suspended problems against the LTM representation of the current item presented to the Recognizer.Working Memory LTMyes 1no 1Figure 2:The Memory ArchitectureThe working memory that emerges has three parts:(1)Fo-cal Store (FS),(2)WM Control unit (the only part of the control unit currently relevant is the Matcher),and (3)Working Store (WS).The Focal Store (FS)holds three items:(1)the current goal of the reasoner (THIS -PROBLEM -CONTEXT ),(2)the current object,artifact or idea being reasoned about (THIS -SKETCHY -SPEC ,which is similar to the P ROBE D ESCRIPTION in Figure 1),and (3)the representation of the current item presented to the Recognizer,module (THIS -SBF-SPEC ),which is retrieved from LTM according to the specification generated by the Recognizer.The working store is more interesting and has four parts.1.A connected graph of related unsolved subproblems,rep-resented as subgoals and the contexts in which they are applicable,called problem contexts (represented as small rectangles in the figure).This graph might be a subset of a problem decomposition stored in LTM when the problem was previously considered,it may have been created dur-ing the reasoning session,or it may be a combination of the two.The goal of the reasoning session is to find a solution for the whole group of related problems.2.Background cues,which provide a history of concepts,de-scriptions,features,and objects that have been considered during reasoning3.A list of Suspended Problems,each represented by a prob-lem context that includes the relevant subgoal,the context in which it is being considered,and the still-incomplete solution description (SKETCHY -SPECS ).More specifically,the representation of suspended subproblems is modeled after the content of problem contexts in design.A problem context in design,and a suspended subproblem in working memory,includes (1)a set of goals and partially ordered constraints that solutions should satisfy;(2)a set of op-tions,or alternatives for achieving those goals 2;and (3)a set of relationships describing how the options satisfy the constraints.These sets are incomplete and contain as much as has been considered so far in addressing the goals.4.A list of Solved Problems,consisting of problem contexts for which solutions (SBF-SPEC s)have been found.These problem contexts have the same structure as do suspended subproblems,but their solution descriptions are complete.This working memory structure,in effect,keeps track of the part of LTM activated during a reasoning session.At most,then,the retention time of working memory is a few hours,requiring only a limited capacity (more work is needed before speculating on how big).The working store accommodates several subproblems (PROBLEM -CTX s),which ideally are related,at the same time.These subproblems are approached one at a time,and if the current one cannot be solved,it is transferred to the list of Suspended Problems.Solved problems are transferred to the Solved Problems queue.A suspended problem is character-ized by a non elaborated specification (SKETCHY -SPEC ),whichFigure3:The Processing Algorithmcannot be used as a successful index in the LTM.A problem is considered solved when the Matcher module recognizes that something in the environment or created on thefly fulfills the requirements formulated in a SKETCHY-SPEC.The whole system is monitored by a global Control module, which is responsible for theflow of problem contexts between working memory,LTM,and the recognizer module,which perceives the world.When a reasoning session ends,the control module makes sure that relevant information from the WM updates the structures in LTM.The Processing AlgorithmMediation between working memory,long-term memory,and perceptual processes are key to working memory’s function-ing.Four control components(see Figure3)are important to using working memory well:(1)The task scheduler loads a graph structure(a set of related subproblems)in the Work-ing Store.(2)The goal-oriented scheduler uses the graph structure and the sets of suspended and pending problems to choose what to do next.Among other things,it suspends subproblems when no headway is being made;it reinstates them when their indexes(specs)are matched.(3)Opportu-nity recognition procedures notice opportunities to reinstate suspended goals and send messages to that effect to the sched-uler.This is accomplished by having perceptual functions (the object recognizer is the only one of these in the scheme presented)focus their attention based on the sketchy specs recorded in suspended subproblems.For example,the ob-ject recognizer seeks to identify objects whose descriptions partially match the sketchy specs associated with suspended subproblems.When such an object is seen,the recognizer asks inference procedures if they can quickly determine if the object has other properties specified in the sketchy spec.If so, the opportunistic component notifies the goal scheduler that a suspended goal ought to be reinstated.(4)Update mech-anisms update the structures in LTM based on recordings in WM.When a new problem is approached,a hierarchical structure is proposed for it.Sometimes the structure is already recorded in memory;sometimes it is on paper;sometimes it must be constructed–we don’t consider that issue right now.As a next step,a small group of related subproblems is brought into focus and loaded into WM.It is essential that this group is keptsmall,because potentially all of its components may becomeactive during reasoning and the computational demand should be limited.Exactly how this choice of subproblems is mademust still be discovered;one option is to bring in only themost connected set of related subproblems and only up to some small threshold.In our example,we assume that the full problem(design ofa quick transportation device)has been considered previouslyand that there are a set of subproblems recorded in memory. In the session we focus on,two subproblems are brought toattention and loaded in WM:the launching device problemand the cushioning material problem.The graph structure in WM has the full problem at the top and these two subproblemshanging off of it as sibling subproblems.One of the subproblems is chosen for focus,and it is loadedinto the Focal Store as the current problem(THIS-PROBLEM-CTX).Here,the launching device problem is chosenfirst. Reasoning procedures work on this problem until it is solved,in which case it is put into the solved problems queue,or untilno progress can be easily made,in which case it is added to the queue of suspended problems.When the need for a collapsible tube emerged in solving the launching device problem,no useful device was recalled from LTM,nor was one seen immediately on the shelves of the store.Thus, this subproblem is suspended.The description created of the collapsible tube(the probe in Figure1)is used as the SKETCHY-SPEC for this suspended problem.When a subproblem is suspended,a new problem is chosen to work on.Here,the cushioning subproblem is selected,and reasoning procedures begin working on it.At the same time,perceptual functions are scanning the en-vironment,looking particularly for things that partially match sketchy specs of suspended problems3.In our case,the object recognizer notices the toilet paper holder on the shelf of thestore.The TPH matches the collapsible tube specification because it is cylindrical and a rigid tube.This is enough of a partial match to the recorded sketchy spec that it asks infer-ence procedures whether the TPH also has variable length.A simple scan of the full representation of a TPH(i.e.,the one in LTM that includes behavioral and structural information) supplies a positive answer(we know that a TPH has to be compressed in order to be assembled to provide support). When a subproblem becomes unblocked due to new infor-mation becoming available,the goal scheduler unblocks the suspended problem and asks reasoning mechanisms to pro-ceed in reasoning about it.This is what happens with the launching device problem.Status and Open IssuesThe working memory model discussed here is implemented as part of the IMPROVISER system(Wills&Kolodner1994a, 1994b).Our original intent was to extend IMPROVISER to allow it to handle and maintain multiple pending problem contexts.However,we soon realized that handling multiple problem contexts was a memory problem and that the mech-anism that would allow that could also be used to explain at least some cases of opportunity recognition.We suspect that this approach will also provide us with ways of explaining for-getting during a long reasoning session and the“freshness”that reasoners feel when coming back to a problem after let-ting it rest for several hours or days.But more exploration is needed before we have good explanation for either of these phenomena.Indeed,we don’t yet have a full explanation of the constraints on memory in handling multiple contexts and in maintaining control of the active goals involved in oppor-tunistic recognition.We do believe,however,that we have proposed a framework within which these questions can be answered quite nicely.We look forward both to continued computational modeling and continued experimentation on people to answer these questions.AcknowledgementsThis research was was funded in part by NSF Grant No.IRI-8921256and in part by ONR Grant No.N00014-92-J-1234. Special thanks to Linda Wills for helpful discussions about the IMPROVISER system.We thank Ashwin Ram,Hari Narayan,Linda Wills and our anonymous reviewers for their comments on this research.ReferencesBarsalou,L.(1992).Cognitive wrence Erl-baum Associates.Birnbaum,L.(1986).Integrated Processing in Planning and Understanding.PhD thesis,Yale University. Birnbaum,L.&Collins,G.(1984).Opportunistic Planning and Freudian Slips.Proceedings of the Sixth Conference of the Cognitive Science wrence Erlbaum Asso-ciates.Dehn,N.(1989).Computer Story-Writing:The role of Re-constructive and Dynamic Memory,PhD thesis,Yale Uni-versity.Grimson,W.E.L.(1990).Object Recognition by Computer: The Role of Geometric Constraints.MIT Press.Hammond,K.,Converse,T.,Marks,M.&Seifert,C.M. (1993).Opportunism and Learning.Machine Learning, 10(pp.279–309).Hayes-Roth,B.&Hayes-Roth,F.(1979).A cognitive model of planning.Cognitive Science3(pp.275–310). Kolodner,J.(1984).Retrieval and Organizational Strategies in Conceptual Memory:A Computer wrence Erlbaum Associates.Kolodner,J.(1993).Case-Based Reasoning.Morgan Kauf-mann(pp.369–382).Kolodner,J.&Wills,L.(1993a).Case-based Creative Design. AAAI Spring Symposium on AI and Creativity.Stanford, CA.March1993.Kolodner,J.&Wills,L.(1993b).Paying Attention to the Right Thing:Issues in Case-Based Creative Design.AAAI Case-Based Reasoning Workshop(pp.19–25).Mayer,N.R.F.(1970).Problem Solving and Creativity:In in-dividuals and Groups.Brooks/Cole Publishing Company, (pp.162–175).Norman,D.A.&Bobrow,D.G.(1979).Descriptions:An intermediate stage in memory retrieval.Cognitive Psy-chology11(pp.107–123).Ram,A.&Hunter,L.(1992).The Use of Explicit Goals for Knowledge to Guide Inference and Learning.Journal of Applied Intelligence,2(1)(pp.47–73).Patalano,A.,Seifert,C.&Hammond,K.(1993).Predictive Encoding:Planning for Opportunities.Proceedings of the Fifteenth Conference of the Cognitive Science Society. Lawrence Erlbaum Associates(pp.800–805).Smith,S.M.&Blankenship,S.E.(1991).Incubation and the persistence offixation in problem solving.American Journal of Psychology,V ol.104,No.1(pp.61–87). Williams,M.&Hollan,J.(1981).The process of retrieval from very long time memory.Cognitive Science5(pp. 87–119).Wills,L.,Kolodner,J.(1994a).Towards More Creative Case-based Design Systems.Proceedings of the Twelfth National Conference on Artificial Intelligence(AAAI-94).Seattle, W A.Wills,L.,Kolodner,J.(1994b).Explaining Serendipitous Recognition in Design.Proceedings of the Sixteenth Con-ference of the Cognitive Science wrence Erl-baum Associates(pp.940–945).。

意大利建筑设计专业院校推荐意大利的建筑专业历史悠久，世界排名领先，深受艺术留学生的青睐。

下面是美行思远小编为大家整理的关于意大利建筑设计专业的相关院校，供大家参考。

意大利建筑设计专业院校推荐1.威尼斯IUAV大学威尼斯IUAV大学坐落于意大利威尼托大区的首府威尼斯自治市，是一所成立于1926年的以建筑研究为主的高等学府。

作为全欧洲最优秀的建筑类大学之一，历年都会出现在由建筑和设计领域内最著名的杂志——《Domus》在全欧洲范围内甄选出的100所顶尖建筑和设计类大学的名单中。

作为意大利唯一一所专门从事所有与人类居住和生活环境有关专业的教学设计与计划编制的大学，它提供了一系列涵盖从建筑设计到产品设计，从城市规划到区域规划，环境研究，剧院设计和绘图、多媒体设计以及通讯等基础设计课程。

同时学校拥有自己的设计公司（IUAV Studi&Progetti-ISP srl），公司由学校教授和毕业生组成，为私人和公共机构进行项目开发设计。

在意大利国内，由CENSIS（意大利社会研究中心）官方每年发布的国内大学建筑系排行榜中，IUAV始终位居前列，最近几年的位次是：2015/2016年为全国第三名，2014/2015年度为全国第三名，其中单项排名中的学术专业水平位居全国首位，2013/2014年度为全国第四名，专业水平同样为全国首位。

2012/2013年度为全国第三名。

威尼斯建筑大学与米兰理工大学、都灵理工大学、费拉拉大学、萨萨里大学这四所大学的建筑系几乎代表了目前意大利建筑专业的最高水平。

相关专业：本科：建筑建造和保护专业、建筑项目的技术和文化专业、城市和区域规划专业硕士：古典和现代建筑专业、建筑项目的技术和文化专业、城市土地和环境规划专业、产品设计专业2.米兰理工大学米兰理工大学创建于1863年，坐落在意大利最富饶的伦巴第大区首府米兰市中心。

这是一所有着悠久历史，师资力量极其雄厚的欧洲顶尖理工大学，也是世界顶尖理工大学之一，在建筑，设计和工程界享有盛名。