US 6,430,731 B1Grant
Methods and apparatus for performing slew dependent signal bounding for signal timing analysis
Issue Date:2002-08-06
•43 Claims
•9 Drawing Sheets
Abstract
Methods and apparatus for use in signal timing analysis with respect to a circuit having at least one gate are provided. In one aspect, the invention includes the step of determining a first constraint slew sensitivity value and a second constraint slew sensitivity value for the at least one gate according to a specified bounding technique. Then, a representative signal for the gate is computed in accordance with the first and second values including an arrival time and slew rate, wherein the representative signal bounds signal paths by bounding a maximum slew sensitivity path and a minimum slew sensitivity path. Such a representative signal may be computed for a worst case late-mode analysis and/or a best case early-mode analysis. The bounding technique may be selected by a user at the time the user inputs the schematic of the circuit on which timing analysis is to be performed. The invention provides for the use of bounding techniques such as, for example, maximum slew, minimum slew, half envelope, full envelope, modified half envelope, modified max slew, modified min slew, least upper bound, and greatest lower bound. The invention may preferably be employed in accordance with static timing analysis associated with VLSI circuit design.
Metadata
Assignee
- International Business Machines Corporation
Inventors
- Jin-Fuw Lee
- Daniel Lawrence Ostapko
- Jeffrey Paul Soreff
- Chak-Kuen Wong
Application Information
Application Number:US 09/368,479
Filing Date:1999-08-04
Priority Date:1999-08-04
Art Unit:7
Classifications
IPC:
G06F 1750
Field of Search:
716 6716 5716 4
Patent Drawings (9 sheets)
Description
Field of the Invention
[0002] The present invention relates generally to signal timing analysis techniques and, more particularly, to methods and apparatus for performing slew dependent signal bounding for use in signal timing analysis.
Background of the Invention
[0003] The propagation of rise and fall times, i.e., slews, has been a long standing problem in static timing analysis, e.g., see J. K. Ousterhout, “A Switch-Level Timing Verifier for Digital MOS VLSI,” in IEEE Trans. on CAD, vol. 3, July 1985, pp. 336-349, and N. Hedenstierna and K. O. Jeppson, “CMOS Circuit Speed and Buffer Optimization,” IEEE Trans. on CAD, vol. 6, March 1987, pp. 270-281. Signals arriving at a given node of a circuit may be generated from different input patterns and may travel through different paths. In a static timing analysis, all of these signals are compared. One signal, namely, the latest signal, is selected to represent the worst-case arrival time bound, and kept to carry through the circuit network (a similar approach can be used for the earliest arriving or early-mode signal). The validity of this approach is based on the observation that the latest signals always yield the latest arrival times when they propagate to the primary output. The monotone nature of signal propagation time in this plain form is true, only if the slew variations among the signals are small.
[0004] Examples of conventional timing analysis approaches which rely on this observation include: R. B. Hitchcock, Sr., “Timing Verification and the Timing Analysis Program,” 19thDesign Automation Conference. June 1982, pp. 594-604; K. A. Sakallah, T. N. Mudge and O. A. Olukotun, “Check Tc and Min Tc: Timing Verification and Optimal Clocking of Synchronous Digital Circuits,” in Proc. ICCAD, November 1990, pp.552-555; G. Szymanski and N. Shenoy, “Verify Clock Schedule,” in Proc. ICCAD, November 1992, pp. 124-131; and J. F. Lee, D. T. Tang, and C. K. Wong, “Algorithm for Circuits with Level-Sensitive Latches,” in IEEE Trans. on CAD, vol. 15, May 1996, pp.535-543.
[0005] However, in reality, circuit designers encounter signals with a wide range of slews. The delay through a CMOS gate depends on the signal slew present at its inputs. The typical approach to accommodate the slew is via the following heuristic method known as the “latest arriving method.” The latest arriving method dictates that during propagation through each gate, always adopt the slew carried by the latest arriving signal. There are two potential pitfalls with this latest arriving method.
[0006] First, the worst-case arrival time bound calculated may be invalid when the latest arriving method is employed, since an earlier signal with a slow slew may eventually reach the primary output later. For example in FIG. 1, assume that the arrival time (a𝑓) of the fast signal at the output of gate 1 is later than that (aS) of the slow signal. Let the delay through gate 2 be d𝑓 and dsrespectively for the fast and slow signals. Since d𝑓<dS, the situation a𝑓+d𝑓<aS+dSmay be reached. This means that the slow signal becomes the latest one at the output of gate 2, and therefore can not be ignored.
[0007] Second, the monotonic property of signal propagation may be violated when the latest arriving method is employed, so that the convergence of the worst-case timing bound is no longer guaranteed. Let us assume that in FIG. 1, the fast signal is only slightly later than the slow signal at the output of gate 1. So the fast signal is chosen to propagate through gate 2, and the worst-case arrival time calculated at the output of gate 2 is a𝑓+d𝑓. If the fast signal is sped up so that its arrival time at the output of gate 1 is a𝑓<aS, then the slow signal is chosen to propagate, and the worst-case time at the output of gate 2 becomes aS+dSwhich is greater than a𝑓+d𝑓. The monotonic property is thus violated.
[0008] Therefore, it would be desirable to have methods and apparatus which provide improved ways to define the worst-case signals for timing analysis purposes so as to, among other things, avoid the problems associated with the above-described conventional latest arriving method.
Summary of the Invention
[0009] The present invention provides methods and apparatus which provide improved ways to define the worstcase signals for timing analysis purposes so as to, among other things, avoid the problems associated with the above-described conventional latest arriving method.
[0010] In one aspect of the invention, a method for use in signal timing analysis with respect to a circuit having at least one gate includes the step of determining a first constraint slew sensitivity value and a second constraint slew sensitivity value for the gate according to a specified bounding technique. Then, a representative signal for the gate is computed in accordance with the first and second values including an arrival time and slew rate, wherein the representative signal bounds signal paths by bounding a maximum slew. sensitivity path and a minimum slew sensitivity path. Such a representative signal may be computed for a worst case late-mode analysis and/or a best case early-mode analysis.
[0011] The bounding technique may be selected by a user at the time the user inputs the schematic of the circuit on which timing analysis is to be performed. The invention provides for the use of bounding techniques such as, for example, maximum slew, minimum slew, half envelope, full envelope, modified half envelope, modified max slew, modified min slew, least upper bound, and greatest lower bound. The invention may be employed in accordance with static timing analysis associated with VLSI circuit design, however, one of ordinary skill in the art will appreciate various other applications such as, for example, circuit tuning given the inventive teachings herein. The invention may also be implemented in the form of a static timer apparatus.
[0012] These and other objects, features and advantages of the present invention will become apparent from the following detailed description of illustrative embodiments thereof, which is to be read in connection with the accompanying drawings.
Brief Description of the Drawings
[0013] FIG. 1 is a diagram illustrating the conventional latest arriving method used to determine the worst-case arrival time bound;
[0014] FIGS. 2A through 2D are diagrams illustrating two signals in several different arrival sequences;
[0015] FIG. 3 is a diagram illustrating several signal bounding methods;
[0016] FIG. 4 is a diagram illustrating arrival times at gate outputs as a function of Δa for the several bounding methods;
[0017] FIGS. 5A and 5B are diagrams of two exemplary cases illustrating dominance according to slew dependent signal bounding techniques of the invention;
[0018] FIG. 6 is a diagram illustrating the quadrants in the s-a plane which contain lower and upper bounds on a signal;
[0019] FIGS. 7A through 7D are diagrams illustrating upper and lower bounds for two signals according to slew dependent signal bounding techniques of the invention;
[0020] FIG. 8 is a diagram illustrating bounding regions for a set of signals according to slew dependent signal bounding techniques of the invention;
[0021] FIG. 9 is a diagram illustrating bounding signals in the s-a plane according to slew dependent signal bounding techniques of the invention;
[0022] FIG. 10 is a diagram illustrating an exemplary implementation of a multiple signal bounding technique according to the invention;
[0023] FIG. 11 is a flow diagram illustrating a static timer implementation of a signal bounding computation process according to the invention;
[0024] FIG. 12 is a block diagram illustrating a hardware implementation of static timer apparatus for a signal bounding computation process according to the invention; and
[0025] FIG. 13 is a diagram illustrating a timing analysis flow using a least upper bound signal process according to the invention.
Detailed Description of Preferred Embodiments
[0026] The present invention is described below in the context of static timing analysis in VLSI (Very Large Scale Integration) circuit design. However, it is to be understood that the present invention is not limited to VLSI circuit design. Rather, the invention is more generally applicable to circuit design in which it is desirable to calculate a worst-case arrival time bound that is valid and for which the monotonic property of signal propagation is not violated.
[0027] It is to be appreciated that the term “processor” as used herein is intended to include any processing device, such as, for example, one that includes a CPU (central processing unit). The term “memory” as used herein is intended to include memory associated with a processor or CPU, such as, for example, RAM, ROM, a fixed memory device (e.g., hard drive), a removable memory device (e.g., diskette), etc. In addition, the term “input/output devices” or “I/O devices” as used herein is intended to include, for example, one or more input devices, e.g., keyboard, for inputting data to the processing unit, and/or one or more output devices, e.g., CRT display and/or printer, for providing results associated with the processing unit. It is also to be understood that various elements associated with a processor may be shared by other processors. Accordingly, software components including instructions or code for performing the methodologies of the invention, as described herein, may be stored in one or more of the associated memory devices (e.g., ROM, fixed or removable memory) and, when ready to be utilized, loaded in part or in whole (e.g., into RAM) and executed by a CPU.
[0028] For ease of reference, the remainder of detailed description will be divided into sections as follows: (I) Characteristics of Slew Dependence; (II) Constructing Bounding Signals for Positive αm; (III) Generating Constraints for Bounding Signals; (IV) Multiple Signal Technique; (V) Static Timer Implementation; (VI) Experimental Results; and (VII) Early-mode Signals.
[0029] I. Characteristics of Slew Dependence
[0030] A VLSI circuit consists of primary inputs (PIs) and outputs (POs), gates, and a network of interconnects. It may be represented by a directed graph G. Nodes of G represent PIs, POs, and pins of gates, while edges of G represent either internal paths from input to output of a gate, or external routing paths between gates. The direction of an edge is defined in such a way that signals flow from its source node to its sink node. The directed graph G may have loops. Each loop must contain some latches or flip-flops, since a loop path consisting of combinational gates will produce a race condition and is not allowed. To perform a static timing analysis on G, worst-case signal arrival times are given at primary input nodes. No input pattern is specified. Under such circumstances, the goal of a static timer is to generate the worst-case arrival time at primary output nodes.
[0031] Signal waveforms traveling on a VLSI chip may be approximated by the ramp function, characterized by four parameters: the voltage levels before and after the transitions, arrival time at the midpoint of the ramp (a), and the duration of the ramp, or slew (s). For digital circuits, the voltage level is either 0 or VDD, and the ramp may represent either a rising or falling transition. For a general signal waveform, the arrival time is defined as the time the signal voltage crosses 0.5 VDD, while the slew is defined as 1.25 times the duration the signal voltage swings between 0.1 VDD and 0.9 VDD. When there is no confusion, the signal waveform defined by w=(s, a) will be called signal as abbreviation for signal waveform.
[0032] For the delay propagation through gates, a static timer may either evaluate the delay on the fly with a fast simulator or calculate the delay with some analytical delay rules. For each input pin to output pin path inside gate i, the signal arrival time and slew at the output pin are functions of the signal arrival time, slew at the input pin and other circuit parameters such as capacitance loading. Consider an internal path from input pin j to the output pin i. Let the arrival time and slew at the input pin be ajand sj, respectively. The arrival time and the slew at the output pin are respectively ajiand sjisuch that:
[0033] aji=aj+𝑓ji(sj) (1)
[0034] sji=gji(sj) (2)
[0035] where 𝑓jiand gjirepresent, respectively, the delay and slew propagation rule for a path from pin j to pin i. These delay rules, defined in equations (1) and (2), may vary from a simple unit delay model, a linear delay model, to a piece-wise quadratic delay model with coefficients extracted from curve-fitting of many simulation runs. The dependence on other circuit parameters was dropped for clarity, as it is irrelevant to an understanding of the present invention.
[0036] According to the invention, we now provide the following definition. For a function 𝑓(s) of slew, the average and instantaneous slew sensitivity are defined as Δ𝑓/Δs and d𝑓/ds. The maximum and minimum sensitivity are respectively defined as
[0037] Both 𝑓jiand gjiare weak functions of slew. By the mean-value theorem of calculus, for any slew interval, there exists a point with the instantaneous sensitivity equal to the average slew sensitivity. Hence, the average slew sensitivities are bounded in the same way as the instantaneous sensitivity:
[0038] αmi≦Δ𝑓ji/Δs≦αMiwhere αmi=Dm𝑓jiand αMi=DM𝑓ji (3)
[0039] βmi≦Δgji/Δs≦βMiwhere βmi=Dmgjiand βMi=DMgji (4)
[0040] From detailed spice simulation for static CMOS gates under different slew (e.g., from 50 ps [pico seconds] to 400 ps) and load conditions (e.g., from 25 ff [fento farads] to 150 ff), the range of slew sensitivities are compiled below in Table 1:
| TABLE 1 | |||
|---|---|---|---|
| Ckt/sensitivity | Δ𝑓/Δs | Δg/Δs | |
| NAND2 | 0.04-0.20 | 0.29-0.36 | |
| NOR2 | 0.05-0.16 | 0.27-0.35 | |
| NAND3 | 0.01-0.22 | 0.20-0.35 | |
| NOR3 | 0.03-0.20 | 0.20-0.36 | |
| AOI22 | 0.06-0.20 | 0.18-0.37 | |
| OAI22 | 0.06-0.20 | 0.18-0.37 | |
[0041] The maximum sensitivities across the CMOS gate library in Table 1 are αM=max{αMi}=0.25 and βM=max{βMi}=0.4. The positive minimum sensitivities, αm=min{αmi} and βm=min{βmi}, from Table 1 imply that both 𝑓jiand gjiare monotonic increasing functions of slew. In the rare case that a slow signal is used to drive a low-threshold dynamic circuit, αmmay become negative, and using the 50% to 50% delay, a slew increase might actually speed up the circuit. Regarding β's, we shall make the following assumption:
[0042] 0<βmi, βMi<1 (5)
[0043] There are some theoretical bounds to α's and β's. Assume that two signals, wiand wj, are present at the same input pin of a gate. Let w′iand w′jbe the corresponding signals at the output pin of the gate. In the following, we shall use b, a, t to denote the 0%, 50% and 100% times of a signal. If wiand wjare non-overlapping in the order ai<ajas shown in FIGS. 2A and 2B, according to the principle of causality, we expect that w′iand w′jare still non-overlapping in the same order a′i<a′j. Mathematically, the condition for non-overlapping signals wiand wjis Δb=bj−bi≧0 and Δt=tj−ti≧0. In terms of Δa=aj−aiand Δs=sj−si, these two relations give Δ(a−s/2)≧0 and Δ(a+s/2)≧0, or Δa≧|Δs/2| in short. Then, the causality condition can be stated as follows: If Δa≧|Δs/2|, then Δa′≧|Δs′|/2. From the gate delay rules, Δa′=Δa+Δ𝑓 and Δs′=Δg where Δ𝑓=𝑓(sj)−𝑓(si) and Δg=g(sj)−g(si). The sufficient condition for Δa′≧|Δs′|/2| can be rewritten as:
[0044] 2(Δa+Δ𝑓)≧|Δg|. (6)
[0045] For the case Δs=sj−si>0 as shown in FIG. 2A, divide equation (6) with Δs and we have 2Δa/Δs+2Δ𝑓/Δs≧|Δg|/Δs. This inequality is true, if the minimum of the left hand side is not smaller than the maximum of the right hand side. The minimum of Δa/|Δs| for non-overlapping signals is ½. The minimum of Δ𝑓/Δs is αm, while the maximum of the right hand side is βM. This means that:
[0046] 1+2αm≧βM (7)
[0047] For the case Δs<0 as shown in FIG. 2B, divide equation (6) with −Δs (a positive number) and we have 2Δa/|Δs|≧2Δ𝑓/Δs+|Δg/Δs|. This inequality is true, if the minimum of the left hand side is not smaller than the maximum of the right hand side. The minimum of Δa/|Δs| for non-overlapping signals is ½. The maximum of Δ𝑓/Δs is αM, while the maximum of |Δg/Δs| is βM. This means that:
[0048] 1≧2αM+βM (8)
[0049] The maximum value for αMthat satisfies equations (5) and (8) is 0.5, while the minimum value for αmthat satisfies equations (5) and (7) is −0.5. Equations (5), (7) and (8) on bounds of α's and β's hold for the sensitivity in Table 1. We assume herein that this is the case. We shall also assume herein that the wiring delay is either negligible, or can be modeled also by equations (5) to (8).
[0050] II. Constructing Bounding Signals for Positive αm
[0051] Now let us consider the typical case with positive αm. In this case, both 𝑓jiand gjiare monotonic increasing functions of slew. In other words, a larger slew at the input produces a longer gate delay, and a larger slew at the output. In the following, we shall use this property to construct a representative signal, when a set of signals, W={w1, w2. . . wn}, are present at the same gate input.
[0052] First let us work with only two signals, wi(slow rise time) and wj(fast rise time). If two signals are non-overlapping, then one signal is always earlier than the other signal, and can be discarded by the principle of causality. Therefore, we need only to consider the overlapping case. If two signals overlap in the lower half voltage interval [0, VDD/2] including VDD/2 as shown in FIG. 2C, we can construct a signal w=(sj, ai) shown as a dashed line. Then, because of a smaller gate delay, after passing one gate, wjwill become earlier than w which is earlier than wiby the amount ai−aj. So wjcan be discarded in this case. We are left with the case that two signals overlap in the upper half voltage interval [VDD/2, VDD] as shown in FIG. 2D. In this case, there are several better bounding methods other than the frequently used latest arriving method, e.g., maximum slew method, full envelope method, and half envelope method. These methods require the construction of a representative signal, which can bound both signals.
[0053] Referring now to FIG. 3, examples of such signal bounding methods are shown along with the late arriving method. The latest arriving method dictates picking the signal wlate=(slate, alate) with the maximum arrival time. For an example, see the solid line between bjand tjin FIG. 3. In the maximum slew method, a signal, wslew=(sslew, aslew), is constructed with the maximum slew and arrival time. For an example, see the dashed-dot line marked “S” in FIG. 3. In the full envelope method, a signal, wfull=(sfull, afull) is constructed which covers all the signals from the right hand side. For an example, see the dotted line marked “F” in FIG. 3. In the half envelope method, a signal, whalf=(shalf, ahalf, is constructed which covers all the signals in the upper half voltage range [VDD/2, VDD] from the right hand side. For an example, see the long dash line marked “H” in FIG. 3.
[0054] For the general case with more than two signals in W, where W={w1, W2, . . . , wi, . . . }, the representative signals are:
[0055]
[0056] In the maximum slew, full envelope and half envelope methods, some new signals, wslew, wfulland whalf, which are worse than any signal in W, are constructed. Therefore, the propagation of these signals will always yield valid upper bounds to the worst-case timing result, while the latest arriving method may lead to an invalid answer. Since only the upper half voltage range is relevant in this positive αmcase, whalfis the tightest upper bound among the three.
[0057] Next we investigate the accuracy penalty paid when we replace a set of signals with a single signal. Again, we shall use the two signal examples in FIG. 2D. At a gate input, we have two non-overlapping signals, wiand wj, with si>sj. Let w′iand w′jbe the corresponding signals at the gate output. Let a=max(ai, aj) and Δa=ai−aj. Then the true worst-case arrival time at the gate output is a′=max(a′i, a′j). For comparison, a′ and the arrival times at output for whalf, wslew, and wfullare plotted in FIG. 4 as functions of Δa for fixed a. Without loss of generality, a is set to 0. A detailed derivation of these arrival times is given below (with Δs=si−sjand Δd=d(si)−d(sj)).
[0058] Referring now to FIG. 4, the four methods are graphically illustrated with respect to output signal time (i.e., the arrival time at gate output) as functions of ai−aj. First, the dependence of a′ on Δa is computed and shown as the piece-wise linear solid line in FIG. 4. There are three regions. For Δa≧0, the slow signal dominates (wj≦wi), and a′=a′i=d(si). For Δa≦−Δd, a′i<a′jand the fast signal dominates: a′=a′j=d(sj). For 0>Δa>−Δd, a′i>a′j(i.e., the earlier slow signal has a later arrival time at the output) and a′=a′i=ai+d(si)=Δa+d(si).
[0059] Next, let we compute the dependence of a′lateon Δa as shown in FIG. 4. For Δa≦0, the fast signal dominates, and a′late=a′j=d(sj). For Δa≧0, the slow signal dominates, and a′slew=a′i=d(si). The maximum error in a′slewoccurs when Δa is slightly negative, that is, Errorlate=−(d(si)−d(sj)).
[0060] Next, we compute the dependence of a′slewon Δa shown as dashed lines in FIG. 4. There are two regions. For the region of Δa≦−Δs/2, the fast signal dominates: a′slew=d(sj). For the region of Δa>−Δs/2, two signals become overlapping, and the slew from the slow signal is selected: a′slew=d(si). The maximum error in a′slewoccurs in the region −Δs/2≦Δa≦−Δd, that is Errorslew=d(si)−d(sj).
[0061] Next, we compute the dependence of a′fullon Δa shown as dotted lines in FIG. 4. For Δa≦−Δs/2, the fast signal dominates: a′full=d(sj). For Δa≧Δs/2, the slow signal dominates: a′full=d(si). For the intermediate region (Δs/2>Δa>−Δs/2), afull=(tl+bj)/2=aj+(ti−tj)/2, sfull=ti−bj=sj+(ti−tj), and a′full=afull+d(sfull)=aj+(ti−tj)/2+(d(sj)+α*(ti−tj))=aj+(½+α)*(ti−tj)+d(sj)=aj+(½α)*(Δa+Δs/2)+d(sj). The maximum error in a′fulloccurs at Δa=−Δd, that is, Errorfull=a′full−d(sj)=(½+α)*(−Δd+Δs/2)=((1−4α2)/4α)(d(si)−d(sj)).
[0062] Next, we compute the dependence of a′halfon Δa shown as heavy solid lines in FIG. 4. There are three regions. For the left region (Δa≦−Δs/2), the fast signal dominates: a′half=d(sj). For the right region (Δa≧0), the slow signal dominates: a′half=d(si). For the middle region (0>Δa>−Δs/2), shalf=2(ti−aj)=si+2Δa and a′half=aj+d(shalf)=d (si+2Δa)≈d(si)+2α*Δa. The maximum error in a′halfoccurs when Δa=−Δd, that is Errorhalf≈d(si)+2α*Δa−d(sj)=(1−2Δ)*(d(si)−d(sj)).
[0063] From these error formulas, we obtain the following ratios:
[0064] Errorhalf/Errorlate=−(1−2α),
[0065] Errorhalf/Errorslew=1−2α,
[0066] Errorhalf/Errorfull=4α/(1+2α),
[0067] For α=0.25 and β=0.4, −0.5*Errorlate=Errorhalf=0.5*Errorslew=0.67*Errorfull. Therefore, the worst-case error in the half envelope method is much smaller than the other methods.
[0068] III. Generating Constraints for Bounding Signals
[0069] The maximum slew, full envelope and half envelope methods described in the previous section require that both the 100% and 50% times be bounded during the propagation through the fan-out paths. The criterion for the bounding signal may be relaxed as follows. Suppose that two signals, wi=(si, ai) and wj=(sj, aj), are present at the same gate input pin. When they propagate through fan-out paths to the latches, if one signal, say wi, arrives at all the catching latches in the fan-out cone always later than the 50% time, wiwill impose stronger timing constraint against the clock edges and shall be considered as the worst-case signal. With this new criterion, we do not require that either 50% or 100% times be bounded in the intermediate stages. It is to be appreciated that the fan-out cone of a gate i is defined as the set of all gates which can be reached from gate i through a sequence of combinational logic gates. In general, the set is a tree rooted at gate i.
[0070] Thus, we provide the following definition. For a pair of signals, wi=(si, ai) and wj=(sj, aj), at the output pin of gate k, let ainand ajnbe the 50% arrival time at latch n on the fan-out cone. Then, we define wi≦wjby the criterion that ain≦ajnfor every latch n on the fan-out cone of gate k. Thus, we say that wiis dominated by wj, or wjdominates wi.
[0071] FIGS. 5A and 5B respectively illustrate two cases where w dominates wi. To explore the sufficient conditions for wi≦w, consider two extreme situations. In the case shown in FIG. 5A, a faster signal w is initially later (a≧aiand s≦si), and the excess delay gain by wiover w through the fan-out propagation is still small enough that the arrival time of w is always later. In the case shown in FIG. 5B, a slower signal w is initially slightly earlier (a≦aiand s≧si), but picks up enough delay gain over widuring the propagation and eventually appears at latches with later arrival times.
[0072] Let us examine more closely the new criterion for a dominating signal. First, a path, which consists of multiple stages of gates, can be regarded as one composite delay element with its own delay rules. The cumulative path delay and the slew of the output signal at the end of path p are functions of the input signal slew s:
[0073] dp=𝑓p(s) and sp=gp(s) (10)
[0074] Let the maximum and minimum slew sensitivity of 𝑓p(gp) be αmp(βmp) and αMp(βMp). In subsection III(a) below, we describe a method to compute these bounds. Assume that a signal w=(s, a) and another signal wi=(si, ai) propagate from the same gate k through a fan-out path p to a capturing latch n. Then the difference in path delays, dp−dip, of two signals are bounded:
[0075] αmp≦(dp−dip)/(s−si)≦αMpwhere αmp=Dm𝑓pand and αMp=DM𝑓p (11)
[0076] Let the arrival times of two signals at latch n be anand ain. Since dp−dip=(an−a)−(ain−ai)=(an−ain)−(a−ai), this leads to:
[0077] (a−ai)/(s−si)+αmp≦(an−ain)/(s−si)≦(a−ai)/(s−si)+αMp. (12)
[0078] For the smaller slew case si≧s in FIG. 5A, a sufficient condition to assure (an−ain)/(s−si)≦0 and hence ain≦anis that the upper bound on the right-hand side of equation (12) is negative or zero. This condition, which is equivalent to −(a−ai)/(s−si)≧αMp, must hold for every fan-out path from gate k to latches. Therefore:
[0079] −(a−ai))/(s−si)≧rMor a+rM* s≧ai+rM*si (13)
[0080]
[0081] where p is any path from gate k to a latch
[0082] Consider the path with maximum slew sensitivity: αMp=rM. Equation (13) implies that the signal with larger a+rM*s will reach the end of the path with a later arrival time an, and hence an is a monotonic increasing function of a+rM*s. We call y=a+rM*s the max-path time of w.
[0083] For the larger slew case si≦s in FIG. 5B, a sufficient condition to assure (an−ain)/(s−si)≧0 and hence ain≦anis that the lower bound on the left hand side of equation (12) is positive or zero. This condition, which is equivalent to −(a−ai)/(s−si)≦αmp, must hold for every fan-out path from gate k to latches. Therefore:
[0084]
[0085] Consider the path with minimum slew sensitivity: αmp=rm. Equation (14) implies that the signal with larger a+rm*s will reach the end of this path with a later arrival time an, and hence anis a monotonic increasing function of a+rm*s. We call x=a+rm*s the min-path time of w.
[0086] Clearly rM≧rm. Signal w dominates wi, if both equations (13) and (14), are satisfied. Hereafter, equations (13) and (14) are respectively called max and min constraint, equations on wi, and rMand rmas the slew sensitivity of max and min constraints. These constraint equations define a feasible region for wi≦w, which is marked as the upper quadrant in FIG. 6. Extension of the two lines representing constraint equations divides the s-a plane into four quadrants as shown in FIG. 6. For signal w in the upper quadrant, wi≦w. For signal w in the lower quadrant of FIG. 6, wiis in the region bounded by the constraint equations on w and hence w≦wi. In either case, we say that two signals are comparable. Signals outside these two quadrants are non-comparable.
[0087] We now consider bound regions of two or more signals. Let us first consider the bounding problem of two signals. Assume that there are two signals, wiand wj, at the same output pin of a gate. Let rmand rMbe the slew sensitivity of the corresponding min and max constraints. We would like to construct an upper-bound signal w=(s, a) and find the condition that w dominates both wiand wj. Also among the upper-bound signals, we would like to find the least upper-bound signal, which is a good candidate for use in the late-mode worst-case timing analysis.
[0088] According to the definition of dominance in the invention, w must satisfy the max and min constraints for both wiand wj, and thus fall into the intersection of upper quadrants of wiand wj. If two signals are comparable, the intersection is equal to the upper quadrant from the dominant signal. Any signal in this quadrant will dominate both signals. The dominant signal, which also resides in this quadrant, is the least upper bounding signal. In FIG. 7B, the fast signal wjis the dominant signal and w must fall into the upper quadrant of wj. In FIG. 7A, the slow signal wiis the dominant signal and w must fall into the upper quadrant of wi. For the non-comparable case in FIG. 7C, the intersection of two upper quadrants can be obtained as follows. First, take the min constraint equation on wjand the max constraint equation on wi. Then, find the intersection, U, of the two lines. Any signal in the upper quadrant of U will dominate both wiand wj, and U is the least upper-bound signal. In the case that rm<0 due to the negative sensitivity, the slopes of lines representing the min constraints become positive as. shown in FIG. 7D. Note that for non-comparable signals, the least upper-bound signal differs from either of the original signals. Thus, the bounding process requires the generation of new signals.
[0089] Now let us consider the general bounding problem involving more than two signals. Assume that there is a set of signals, W={w1, w2, . . . wi} at the same output pin of a gate. Let rmand rMbe the sensitivity of the corresponding min and max constraints. The bounding problem is to find, for the late-mode worst-case timing, the least upper-bound signal which dominates all signals in W, and whereas, for the early-mode best-case timing, the greatest lower-bound signal which is dominated by all signals in W.
[0090] In FIG. 8, the lines are drawn to represent the min and max constraints for each signal, w1, w2, . . . wi. The lines for min constraints having the same slope are parallel to each other, and so are the lines for max constraints. The min and max path time of wiis xi=ai+rm*si, and yi=ai+rM*si. The min and max path time of an upper-bound signal w=(s,a) must satisfy x=a+rm*s≧xiand y=a+rM*s≧yifor every i. Hence, x≧max (x1, x2, . . . xn) and y≧max (y1, y2, . . . yn). Let U=(sU, aU) be the intersection point: xU=max(x1, x2, . . . xn) and yU=max(y1, y2, . . . yn). Then, xU≧xiand yU≧yi. U dominates wiand is an upper-bound signal. On the other hand, any bounding signal satisfies x≦xUand y≧yU. U is indeed the least upper-bound signal.
[0091] By similar analysis, the greatest lower-bound signal L can be derived for use in the early mode timing analysis, and thus we can state the following:
[0092] Proposition 1: For a set of signals, W={w1, w2, . . . wn}, let xi=ai+rm*si, and yi=ai+rM*si. The least upper-bound signal U=(sU, aU) is defined by aU+rm*sU=max(x1, x2, . . . xn) and aU+rM*sU=max(y1, y2, yn), whereas the greatest lower bound signal L=(sL, aL) is defined by aL+rm*sL=min(x1, x2, . . . xn) and aL+rM*sL=min(y1, y2, . . . yn).
[0093] (a) Computations for rmand rM
[0094] Assume that signal, w=(s, a), is present at the output pin of gate i. Consider a path p from this output pin to a latch n. Let j be a gate in the fan-out of gate i, and q be the sub-path from the output pin of j to n. Let the signal at the output pin of gate j and the input pin of latch n be wj=(sj, aj) and wn=(sn, an). Then from the delay rules of gate j, we have dj≡aj−a=𝑓j(s) and sj=gj(s). From the delay rules of path q, we have dq≡an−aj=𝑓q(sj) and sn=gq(sj). The delay rules for path p can be defined by the following recursive formula:
[0095] gp(s)=(gq∘gj)(s)
[0096] 𝑓p(s)=an−a=(𝑓j+𝑓q∘gj)(s) (15)
[0097] where ∘ is the chaining operator: 𝑓∘g(s)=𝑓(g(s)). We then state the following:
[0098] Proposition 2: Assume that DMg≧Dmg≧0. We have the following bounds on the slew sensitivity for composite functions, 𝑓+g and 𝑓∘g:
[0099] (a) DM(𝑓+g)≦DM𝑓+DMg and Dm(𝑓+g)≧Dm𝑓+Dmg
[0100] (b) DM(𝑓∘g)≦max{DM𝑓*DMg, DM𝑓*Dmg} and Dm(𝑓∘g)≧min{Dm𝑓*Dmg, Dm𝑓*DMg}.
[0101] Proposition 2 may be proven in the following manner. The slew sensitivity of 𝑓+g is d(𝑓+g)(s))/ds=𝑓′(s)+g′(s) where ′ is the differentiation operator. Since the maximum point of 𝑓+g does not maximize 𝑓 and g separately in general, we have DM(𝑓+g)≦DM𝑓+DMg. Similarly, the lower bound Dm(𝑓+g)≧Dm𝑓+Dmg can be proven. The slew sensitivity of 𝑓∘g, can be obtained from the chain rules as d𝑓(g(s))/ds=𝑓′(g(s))*g′(s). Let smand sMbe the slew which minimizes and maximizes g′(s): Dmg=g′(sm), DMg=g′(sM). By definition, Dm𝑓≦𝑓′(g(s))≦DM𝑓, and 0≦Dmg≦g′(s)≦DMg since all β's are assumed to be positive. Since DM𝑓 and Dm𝑓 may be negative, min(Dm𝑓*DMg, Dm𝑓*Dmg)≦𝑓′(g(s))*g′(s)≦max(DM𝑓*DMg, DM𝑓*Dmg). If the sign of DM𝑓 is positive, DMg*DM𝑓 is an upper bound to DM(𝑓∘g). Otherwise, DM𝑓*Dmg is an upper bound.
[0102] Applying Proposition 2 to equation (15), we obtain DM𝑓p≦max{αMj+DM𝑓q*βMj, αMj+DM𝑓q*βmj} where αmjDm𝑓j, αMj=DM𝑓j, βMj=DMgj, and βmj=Dmgj. Since the set of fan-out paths from gate i is the union of fan-out paths from all its immediate successor gates j, we have
[0103] The same analysis applies to rmi. We obtain the recursive relations:
[0104]
[0105] This leads to the following algorithm to compute the slew sensitivity of constraints referred to herein as Algorithm trace_sensitivity:
[0106] (1) Sort the gate in the reverse order. That is, if there is a path from gate i to gate j, then place j before i.
[0107] (2) Initialize rmiand rMito 0 at latches. /*start the backward propagation through gates*/
[0108] (3) For each gate i, visit the set of successor gates {j} and compute rMiand rmias follows:
[0109] (a) If rMjis positive, set rMij=αMj+rMj*βMj. Otherwise, set rMij=αMj+rMj+Mj*βmj.
[0110] (b) If rmjis positive, set rmij=αmj+rmj*βmj. Otherwise, set rmij=αmj+rmj*βMj.
[0111] (c) Compute rMiand as the maximum of {rMij} and rmias the minimum of {rmij}.
[0112] Then we can start a forward propagation pass to compute the optimum arrival times using Proposition 1. The cost of this backward sweep algorithm is linear in the number of gates in the circuits.
[0113] We now deal with the case of positive α. For the case that α's are all positive, the slew sensitivities of constraints are also positive, and we have:
[0114]
[0115] Since βMk≦βMand αMk≦αMfor all gates in the fan-out cone, rMfor max constraint is bounded above by the series
[0116] where N is the maximum number of stages in the fan-out cone. Since the geometric series converges very fast, we can set N in equation (19) to infinity for computing the upper bound:
[0117] rM≦αM/(1−βM) (20)
[0118] Since βmk≧βmand αmk≧αmfor all gates in the fan-out cone, rmfor min constraint is bounded below by the series
[0119] where n is the minimum number of stages in the fan-out cone. For signals at latches, n=0 and rm=0. For signals on path with at least one gate, n>0 and:
[0120] rm≧αm (22)
[0121] For incomplete designs, in which the maximum (minimum) number of stages are unknown, equations (20) and (22) can be used for the max and min constraint.
[0122] (b) Methods Characterized by Two Slopes
[0123] Referring to FIG. 9, bounding signals in s-a plane are shown. It is to be appreciated that in sections III(b), III(c), and IV to VI, we focus on bounding signal methods for the worst case late-mode analysis, and then, in section VII, we focus on bounding signal methods for the best case early-mode analysis. The shaded area in FIG. 9 shows the feasible quadrant for the upper bound of two non-comparable signals, wiand wj(si≧sj). For any point w=(s, a) in the s-a plane, we can draw two lines, one through wjand the other one through wi. Let the slopes of these lines be −r1 and −r2. Then:
[0124] a+r1*s=uj=aj+r1*sjand a+r2*s=vi=ai+r2*si (23)
[0125] On the other hand, given r1 and r2, we can solve equation (23) for a and s:
[0126] s=(vi−uj)/(r2−r1)=(ai−aj+r2*si−r*1sj)/(r2−r1)
[0127] a=(r2*uj−r1*vi)/(r2−r1)=(−r1*ai+r2*aj−r1*r2*(si−sj))/(r2−r1) (24)
[0128] Given r1 and r2, equation (24) provides a formula for constructing a representative signal from two signals. Not all values are good for r1 and r2. If we want the signal to reside in the feasible quadrant within the slew range [sj, si], then the slopes have to satisfy the following bounds:
[0129] w(from equation (24))≧wi, wj, if r2≧rM≧rm≧r1 (25)
[0130] The construction methods with two slopes can be extended to the general case when there is a set of signals, W={w1, w2, . . . wi} present at the same gate output. First, a representative signal w12is constructed for a pair of signals, say the first two. Then, a representative signal w123for a triplet of signals is constructed with w12and a third signal. This process can be continued recursively, and leads to the following generalization of equation (23) for the representative signal (s,a):
[0131] a+r1*s=max(u1, u2, . . . un) and a+r2*s=max(v1, v2, . . . vn),
[0132] where uk=ak+r1*sk, and vk=ak+r2*skfor k=1, 2, . . . n. (26)
[0133] The following proposition can then be stated:
[0134] Proposition 3: In the general case with a set of signals W, if r2≧rM≧rm≧r1, then the signal point defined by equations (24) and (26) is inside the feasible quadrant formed by constraints rMand rm.
[0135] This can be proven in the following manner. Let wibe the signal with maximum a+rm*s and wjbe the signal with maximum a+rM*s. From wiand wj, we construct two signals: U=(sU, aU) at the intersection of lines of slopes −rmand −rM, and wk=(sk, ak) at the intersection of lines of slopes −r1 and −r2. According to equation (25), wk≧U and thus satisfies the min and max constraint equations: ak+rm*sk≧xUand ak+rM*sk≧yUwhere xU+aU+rm*sUand yU=aU+rM*sU. With substitution from equation (24), the min constraint is transformed to (r2*uj−r1*vi+rm*(vi−uj))/(r2−r1)≧xU, or equivalently (r2−rm)*uj+(rm−r1)*vi≧xU*(r2−r1). Let w=(s, a) be the signal constructed with equation (26). Since u=a+r1*s≧ujand v=a+r1*s≧vi, we have (r2−rm)*u+(rm−r1)*v≦(r2−rm)*uj+(rm−r1)*vi≧xU*(r2−r1), and w also satisfies the min constraint. Similarly, it can be proven that w satisfies the max constraint. Therefore, w is inside the upper quadrant of U. According to Proposition 1, w is feasible.
[0136] Methods from Section II for positive α can all fit into this characterization scheme. The corresponding values for r1 and r2 are listed in Table 2 below. Point S (the maximum slew method) with r1=0 and r2=∞ is definitely inside the feasible quadrant. The latest selection method picks the fast-rise later signal wj, which is outside the feasible quadrant. H and F will fall in the feasible quadrant, if r2=0.5≧rM. From equation (20), this translates into a sufficient condition: 1≧2αM+βM, which is one of our assumptions, i.e., equation (8).
[0137] Four new points, M, MS, mS and U, are shown in Table 2. M has the same arrival time as S, but with slew less than both H and S. MS has the maximum slew and an arrival time earlier than S. mS has the minimum slew and an arrival time later than S. U is the optimal upper bound signal.
[0138] It can be proven that equations (24) and (26) can be reduced to equation (9) when appropriate r1 and r2 are substituted. For example, for the half envelope method (r1=0, r2=0.5), xi=ai, yi=ai+0.5*si=ti, and the signal with maximum aiand maximum tiis selected.
[0139] It is to be appreciated that with respect to methods in Table 2: If rmbecomes negative due to negative α's, then points M, H, and S become infeasible, and this leaves U, F, MS and mS as the only valid methods for worst-case timing analysis.
| TABLE 2 | |||
|---|---|---|---|
| Methods | r1 | r2 | |
[0140] (c) Monotonicity
[0141] Monotonic delay is often a desired property which can be used to speed up the circuit timing and tuning process. According to the invention, we present the following definition for monotonicity:
[0142] For a set of signals, W={w1, w2, . . . wn}, let wibe the one with maximum a+r1*s and wjbe the one with maximum a+r2*s. Then s and a of the representative signal from equations (24) and (26) are functions of four variables ai, aj, si, sj, and so are the min and max path times: x=a+rm*s and y=a+rM*s. Since the arrival times at latches are monotonic increasing functions of the min and max path times, we shall say the representative signal construction method preserves the monotonic property, if x and y are monotonic increasing with all four variables ai, aj, si, sj.
[0143] Therefore, to satisfy the monotonic property, the partial derivatives of min and max path times with respect to ai, aj, si, sj, need to be either zero or positive. In other words, the 4-dimensional gradient vectors ∇x=(∂x/∂ai, ∂x/∂aj, ∂x/∂si, ∂x/∂sj) and ∇y=(∂y/∂ai, ∂y/∂aj, ∂y/∂si, ∂y/∂/sj) must not contain any negative derivative. From equation (24), we have ∇s=(1, −1, r2, −r1)/(r2−r1) and ∇a=(−r1, r2, −r1*r2, r1*r2)/(r2−r1) for the case of constant r1 and r2. This leads to:
[0144] ∇x=(rm−r1, r2−rm, r2*(rm−r1), r1*(r2−rm))/(r2−r1)
[0145] ∇y=(rMr1, r2−rM, r2*(rM−r1), r1*(r2−rM))/(r2−r1) (27)
[0146] Under the assumptions that r2≧rM≧rm≧r1, the derivatives of x and y with respect to ai, ajare never negative, while the sign of derivatives with respect to si(sj) are the same as that of r2(r1). If r1>0, then there are no negative derivatives. Thus we have proven the following proposition:
[0147] Proposition 4: The method using two constant slopes r1 and r2, such that r2≧rM≧rm≧r1≧0, preserves the monotonic property.
[0148] The latest arriving method is not monotonic as described in the background section. Because of negative r1, points F and mS are also not monotonic. For the case that rm≧0, points S, H, M, MS and U preserves the monotonic property. For the case that rM≧0 and rm≦0, points S, H and M preserve the monotonic property but are not feasible, while points U, F, mS and MS do not preserve the monotonic property but are feasible. For the case that rM≦0 and rm≦0, only the infeasible point S preserves the monotonic property.
[0149] IV. Multiple Signal Technique
[0150] Where further accuracy is required, the present invention provides a multiple signal approach. The multiple signal approach is described in the context of FIG. 10. For example, in FIG. 10, three fan-out gates have different slew-dependent rules, d1, d2, d3. The arrival times at these outputs, a′1, a′2, a′3, for three signals w1, w2, W3, are listed in the first three columns of Table 3 below. At output a′1, w1produces the latest signal. At output a′2, w2produces the latest signal. At output a′3, w3produces the latest signal. No single signal can be chosen to generate the latest signals at all outputs. The arrival time from several bound methods are also listed in Table 3. No bound method gives the correct latest arrival time for all three outputs. The least upper bound method U predicts the correct a′1and a′3, but the wrong a′2. Therefore, to compute the true latest arrival times at all latches, we need to store and propagate a set of signals, which contain the combinations of all possible slew and arrival times.
| TABLE 3 | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| w | w | w | ||||||||
| (s | (s | (s | ||||||||
| a | a | a | U | M | H | F | S | mS | MS | |
[0151] However, the signal set may grow very large in size, as it propagates through a large circuit network. Advantageously, we can use the dominance relation ≧ to trim out those signals, which are dominated by other signals. In other words, we need to consider only maximal signals. A maximal signal is defined as one which no other signal dominates, and a maximal signal set is defined as the subset consisting of maximal signals. Since the dominance order of signals will not change during the propagation through the circuit, the worst-case timing can be computed using waveforms represented by the maximal signal set. The maximal signal set can be obtained by a simple sorting, according to Proposition 5 below. An example of sorted maximal signal set is shown in FIG. 8.
[0152] Proposition 5: For a maximal signal set, W={w1, w2, . . . wn}, let xi=ai+rm*siand yi=ai+rM*si. Then:
[0153] (1) All xi's and yi's are distinct.
[0154] (2) If {xi} is sorted in a strictly descending sequence, then {yi} will be sorted simultaneously in a strictly ascending sequence.
[0155] This may be proven in the following manner. For the first part, let us assume that xi=xj. If yi>yj, then wjis not maximal. If yi<yj, then wiis not maximal. Both lead to contradiction. So xi≠xj. In a similar way, we can prove that all yi's are distinct. For the second part, we observed that if xi>xjand yi>yj, then wi>wj. Therefore, xi's and yi's are in the opposite order.
[0156] The trimming mechanism by the maximal set is very powerful as is illustrated in the experiment discussed in Section VI. According to the principle of causality, one of two non-overlapping signals will dominate the other. So the maximal signal set will contain only overlapping signals. If r's and the maximal set are not available, we can use the set of overlapping signals instead in the signal propagation.
[0157] V. Static Timer Implementation
[0158] Referring now to FIG. 11, a flow diagram of a static timer implementation of a signal bounding computation process according to the invention is shown. An exemplary hardware implementation of a static timer employing bounding signal computation according to the invention will be given in the context of FIG. 12.
[0159] First, in step 100 of FIG. 11, the circuit schematics of the VLSI chip on which static timing analysis is to be performed are input to the static timer. The schematics include the gates, latches and paths associated with the circuit and may be input in any conventional manner. In step 110, the gates of the circuit are levelized. That is, the gates are sorted in order and assigned levels. Next, in step 120, a user selects the method that is preferred for the particular input circuit, e.g., a method from Table 2 above. The user's method selection is preferably entered at the same time the circuit schematic is specified. Given the user's selection, the slew sensitivity of the maximum and minimum constraints is computed in accordance with the Algorithm trace_sensitivity above and Table 2.
[0160] In step 140, signals are propagated from primary inputs through wires to inputs of level-1 gates. A gate is then taken from the ordered list, in step 150. Signals representing the arrival time and slew are computed at the gate output according to equations (1) and (2) above. Then, in step 160, the representative signal (s, a) is computed with the arrival time and slew according to equation (26) above. The signals computed in steps 150 and 160 are then propagated through connecting wires to the input pins of fan-out gates, in step 170, adding the net delay if significant.
[0161] In step 180, it is determined if there are any more gates in the circuit. If there are more gates, steps 150 through 170 are iteratively repeated for each gate. If all gates in the input schematic have been processed, then a timing report representing the results of the static timing analysis is output to the user in step 190.
[0162] Referring now to FIG. 12, an exemplary hardware implementation of static timer apparatus employing bounding signal computation according to the invention is shown. In this implementation, a static timer system includes a processor 210, memory 220 and I/O devices 230. Processor 210 performs the steps of the methodologies of the invention described herein, e.g., the illustrative bounding signal computation described in FIG. 11. The processor 210 employs memory 220 in performing such operations, e.g., storing user input selections and circuit schematics, etc. Memory 220 may also store software code for implementing the methodologies for execution by the processor 210. I/O devices 230 may include a keyboard, mouse, etc., for allowing a user to input data for use by the invention and a display, printer, etc.; for presenting a user with results associated with the processor operations, e.g., timing report, etc.
[0163] Referring now to FIG. 13, a diagram is shown illustrating a timing analysis flow using a least upper bound signal process according to the invention. The example follows the steps of the illustrative process of FIG. 11. It is to be appreciated that subscript U will be used in this example to denote the arrival time and slew computed with the least upper bound method. For simplicity, wire delay is assumed to be zero in this example.
[0164] First, the user specifies the schematic of FIG. 13 as the circuit schematic and the least upper bound method as the desired method (steps 100 and 120). The schematic includes gates 0, 1, 2 and 3. The gates are sorted in the following order: 0, 1, 2, 3 (step 110). Gate 0 is at level 1, while gates 1, 2 and 3 are at level 2. Then, following the reverse order of gates, the Algorithm trace_sensitivity is performed to compute rmand rMfor each gate (step 130). So, in this example, gates 1, 2 and 3 have rm=0 and rM=0, and gate 0 has rm=0.15 and rM=0.25. Then, we set r1 to rmand r2 to rMaccording to Table 2, since the least upper bound is the method the user selected.
[0165] Next the signals denoted by s1i, a1i, and s2i, a2i, and s3i, a3iare propagated from primary inputs to the inputs of gate 0 (step 140). The corresponding three signals (s1, a1), (s2, a2), (s3, a3) are then computed at the output of gate 0 (step 150). From these three signals, the bounding signal is computed (step 160) by solving equation (26) to yield (sU=80, aU=10). The bounding signal (sU, aU) is then propagated to the inputs of gates 1, 2 and 3 (step 170). The signals are propagated through gate 1 in the same way (step 180). Since only one signal (sU, aU) remains at the input of gate 1, it is the one to propagate to the next stage, which in this example is a primary output. The signals are also propagated through gates 2 and 3 in the same way (step 180). Lastly, the bounding signals are reported at the primary outputs (step 190).
[0166] VI. Experimental Results
[0167] To assess the accuracy of these different methods, the following experiments are performed on ISCAS benchmark circuits. Linear delay rules are used: gate delay=100 ps+0.25*(Tx−200 ps)+100 ps*(gain−1), and gate output slew=200 ps+0.4*(Tx−200 ps)+200 ps*(gain−1), where Tx is the input slew and gain is the ratio of capacitance load over the input capacitance. For simplicity, we assume that all gates have the same size, and the gain is equal to the number of fan-outs. Sharp signals with zero slew are fed into primary inputs and latches at time 0. Because of the slew-dependent delay rule, the signal slew will gradually increase and spread into a range between 0 and some maximum slew value after propagating through gates.
[0168] Six timing runs are done: the exact multiple signal method (E), the latest arriving method (late), the max slew method (S), the full envelope method (F), the half envelope method (H), and the upper bound method (U). Table 4 lists the maximum difference of worst-case delay at primary outputs between the last five methods and the exact method. The latest arriving method gives too optimistic timing in the case of circuit c432. The maximum slew method is the most pessimistic method among the five. The full envelope method sometimes yields modest error. The half envelope method performs better than the full envelope method, while the upper.bound method produces the best result among the five. In this experiment, the number of signals in the maximal signal set never exceeds 10, because of the trimming. Hence, the exact method with multiple signals is also be a viable solution. In the real circuit with different gate sizes and the range of slew larger, we may encounter situations that the slew has a larger effect on the timing.
| TABLE 4 | |||||
|---|---|---|---|---|---|
| Latest | max | Full | Half | Upper | |
| arriving | slew | envelope | envelope | bound | |
[0169] VII. Early-mode Signals
[0170] The best case early-mode signals may be obtained in a similar fashion to the late-mode signal. Let us consider the example in FIG. 9 of two non-comparable signals, wiand wj(si≧sj). For any point w=(s, a) in the feasible region for the early signals, i.e., the lower quadrant of the s-a plane, we can draw two lines, one through wiand the other one through wj. Let the slope of the line drawn between wiand w be −r1; and the slope of the line drawn between wjand w be −r2. Then, the signal w is given by:
[0171] a+r1*s=ui=ai+r1*siand a+r2*s=vj=aj+r2*sj (28)
[0172] In the general case when there is a set of signals, W={w1, w2, . . . wi} present at the same gate output, a similar argument used for the late-mode signals can be applied here. From equation (26), we obtain the following corresponding equation for the representative early-mode signal:
[0173] a+r1*s=min(u1, u2, . . . un) and a+r2*s=min(v1, v2, . . . vn),
[0174] where ukak+r1*sk, and vk=ak+r2*skfor k=1, 2, . . . n. (29)
[0175] The corresponding methods for early-mode signals are i for the traditional early arriving method, S′ for the minimum slew method, H′ for the half envelope method, F′ for the full envelope method, M′ for the modified half envelope method, MS′ for the modified minimum slew method, mS′ for the modified maximum slew method, and L for the greatest lower bound method. These early signals are marked in FIG. 9 and the corresponding slopes, r1 and r2 are listed in Table 5 below.
| TABLE 5 | |||
|---|---|---|---|
| Methods | r1 | r2 | |
[0176] It is to be appreciated that a static timer implementation such as is illustrated in FIG. 11 may also include early-mode analysis in accordance with Table 5 and equation (29). As such, the steps of the method shown in FIG. 11 in which early-mode analysis is added parenthetically indicate the early-mode analysis portion, i.e., steps 130 and 160.
[0177] Although illustrative embodiments of the present invention have been described herein with reference to the accompanying drawings, it is to be understood that the invention is not limited to those precise embodiments, and that various other changes and modifications may be affected therein by one skilled in the art without departing from the scope or spirit of the invention.
Claims
What is claimed is:
1. A method for use in signal timing analysis with respect to a circuit having at least one gate, the method comprising the steps of:
determining a first constraint slew sensitivity value (r1) and a second constraint slew sensitivity value (r2) for the gate according to a specified bounding technique; and
computing a representative signal for the gate in accordance with the first and second values including an arrival time and slew rate, wherein the representative signal bounds signal paths by bounding a maximum slew sensitivity path and a minimum slew sensitivity path.
2. The method of claim 1, wherein the bounding technique is a half envelope technique wherein r1 is about zero and r2 is about 0.5.
3. The method of claim 1, wherein the bounding technique is a full envelope technique wherein r1 is about −0.5 and r2 is about 0.5.
4. The method of claim 1, wherein the bounding technique is a modified half envelope technique wherein r1 is about zero and r2 is a maximum constraint slew sensitivity value.
5. The method of claim 1, wherein the computed signal is representative of a worst case arrival time and slew rate.
6. The method of claim 5, wherein the bounding technique is a maximum slew technique wherein r1 is about zero and r2 is substantially infinite.
7. The method of claim 5, wherein the bounding technique is a modified maximum slew technique wherein r1 is a minimum constraint slew sensitivity value and r2 is substantially infinite.
8. The method of claim 5, wherein the bounding technique is a modified minimum slew technique wherein r1 is substantially negatively infinite and r2 is a maximum constraint slew sensitivity value.
9. The method of claim 5, wherein the bounding technique is a least upper bound technique wherein r1 is a minimum constraint slew sensitivity value and r2 is a maximum constraint slew sensitivity value.
10. The method of claim 1, wherein the computed signal is representative of a best case arrival time and slew rate.
11. The method of claim 10, wherein the bounding technique is a minimum slew technique wherein r1 is about zero and r2 is substantially infinite.
12. The method of claim 10, wherein the bounding technique is a modified minimum slew technique wherein r1 is a minimum constraint slew sensitivity value and r2 is substantially infinite.
13. The method of claim 10, wherein the bounding technique is a modified maximum slew technique wherein r1 is substantially negatively infinite and r2 is a maximum constraint slew sensitivity value.
14. The method of claim 10, wherein the bounding technique is a greatest lower bound technique wherein r1 is a minimum constraint slew sensitivity value and r2 is a maximum constraint slew sensitivity value.
15. The method of claim 1, wherein the circuit includes a plurality of gates.
16. The method of claim 15, further including the step of assigning levels to the plurality of gates prior to determining the first constraint slew sensitivity and second constraint slew sensitivity values for each of the plurality of gates.
17. The method of claim 16, further including the step of propagating signals from inputs of the circuit to an input of the first gate at the first level, each signal representing an arrival time and slew rate.
18. The method of claim 17, further including the step of computing signals at an output of the first gate at the first level, the signals corresponding to respective gates at a next level to which the first gate at the first level is connected, and each signal representing an arrival time and slew rate.
19. The method of claim 18, wherein the representative bounding signal is computed from the signals computed at the output of the first gate at the first level.
20. The method of claim 19, wherein a representative bounding signal is computed for each gate at each level.
21. The method of claim 20, wherein the representative bounding signals for each gate are presented to a user in a timing report.
22. Apparatus for use in signal timing analysis with respect to a circuit having at least one gate, the apparatus comprising:
at least one processor operative to determine a first constraint slew sensitivity value and a second constraint slew sensitivity value for the at least one gate according to a specified bounding technique, and to compute a representative signal for the gate in accordance with the first and second values including an arrival time and slew rate, wherein the representative signal bounds signal paths by bounding a maximum slew sensitivity path and a minimum slew sensitivity path; and
memory coupled to the at least one processor which stores one or more of the computed values or signals.
23. The apparatus of claim 22, wherein the bounding technique is a half envelope technique wherein r1 is about zero and r2 is about 0.5.
24. The apparatus of claim 22, wherein the bounding technique is a full envelope technique wherein r1 is about −0.5 and r2 is about 0.5.
25. The apparatus of claim 22, wherein the bounding technique is a modified half envelope technique wherein r1 is about zero and r2 is a maximum constraint slew sensitivity value.
26. The apparatus of claim 22, wherein the computed signal is representative of a worst case arrival time and slew rate.
27. The apparatus of claim 26, wherein the bounding technique is a maximum slew technique wherein r1 is about zero and r2 is substantially infinite.
28. The apparatus of claim 26, wherein the bounding technique is a modified maximum slew technique wherein r1 is a minimum constraint slew sensitivity value and r2 is substantially infinite.
29. The apparatus of claim 26, wherein the bounding technique is a modified minimum slew technique wherein r1 is substantially negatively infinite and r2 is a maximum constraint slew sensitivity value.
30. The apparatus of claim 26, wherein the bounding technique is a least upper bound technique wherein r1 is a minimum constraint slew sensitivity value and r2 is a maximum constraint slew sensitivity value.
31. The apparatus of claim 22, wherein the computed signal is representative of a best case arrival time and slew rate.
32. The apparatus of claim 31, wherein the bounding technique is a minimum slew technique wherein r1 is about zero and r2 is substantially infinite.
33. The apparatus of claim 31, wherein the bounding technique is a modified minimum slew technique wherein r1 is a minimum constraint slew sensitivity value and r2 is substantially infinite.
34. The apparatus of claim 31, wherein the bounding technique is a modified maximum slew technique wherein r1 is substantially negatively infinite and r2 is a maximum constraint slew sensitivity value.
35. The apparatus of claim 31, wherein the bounding technique is a greatest lower bound technique wherein r1 is a minimum constraint slew sensitivity value and r2 is a maximum constraint slew sensitivity value.
36. The apparatus of claim 22, wherein the circuit includes a plurality of gates.
37. The apparatus of claim 36, wherein the processor is further operative to assign levels to the plurality of gates prior to determining the first constraint slew sensitivity and second constraint slew sensitivity values for each of the plurality of gates.
38. The apparatus of claim 37, wherein the processor is further operative to propagate signals from inputs of the circuit to an input of the first gate at the first level, each signal representing an arrival time and slew rate.
39. The apparatus of claim 38, wherein the processor is further operative to compute signals at an output of the first gate at the first level, the signals corresponding to respective gates at a next level to which the first gate at the first level is connected, and each signal representing an arrival time and slew rate.
40. The apparatus of claim 39, wherein the representative bounding signal is computed from the signals computed at the output of the first gate at the first level.
41. The apparatus of claim 40, wherein a representative bounding signal is computed for each gate at each level.
42. The apparatus of claim 41, wherein the representative bounding signals for each gate are presented to a user in a timing report.
43. An article of manufacture for use in signal timing analysis with respect to a circuit having at least one gate, comprising a machine readable medium containing one or more programs which when executed implement the steps of:
determining a first constraint slew sensitivity value (r1) and a second constraint slew sensitivity value (r2) for the gate according to a specified bounding technique; and
computing a representative signal for the gate in accordance with the first and second values including an arrival time and slew rate, wherein the representative signal bounds signal paths by bounding a maximum slew sensitivity path and a minimum slew sensitivity path.
Patent Citations (5)
Non-Patent Literature (9)
- Venkatesh et al, “Timing Abstraction of Intellectual Property Blocks,” IEEE, May 1997, pp. 99-102.Applicant
- Kayssi et al, “Analytical Transitient Response of CMOS Inverters,” IEEE, Jan. 1992, pp. 42-45.Applicant
- Xiao et al, “Worst Delay Estimation in Crosstalk Aware Static Timing Analysis,” IEEE, Sep. 2000, pp. 115-120.Applicant
- R.B. Hitchcock, Sr. et al., “Timing Analysis of Computer Hardware,” IBM J. Res. Develop., vol. 26, No. 1, pp. 100-105, Jan. 1982.Applicant
- J.K. Ousterhout, “A Switch-Level Timing Verifier for Digital MOS VLSI,” IEEE Transactions on Computer-Aided Design, vol. CAD-4, No. 3, pp. 336-348, Jul. 1985.Applicant
- N. Hedenstierna et al., “CMOS Circuit Speed and Buffer Optimization,” IEEE Transactions on Computer-Aided Design, vol. CAD-6, No. 2, pp. 270-281, Mar. 1987.Applicant
- K.A. Sakallah, et al., “checkTc and minTc: Timing Verification and Optimal Clocking of Synchronous Digital Circuits,” Proc. ICCAD, pp. 552-555, Nov. 1990.Applicant
- T.G. Szymanski et al., “Verifying Clock Schedules,” Proc. ICCAD, pp. 124-131, Nov. 1992.Applicant
- J-F Lee et al., “A Timing Analysis Algorithm for Circuits with Level-Sensitive Latches,” IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 15., No. 5, pp. 535-543, May 1996.Applicant