US 6,499,131 B1Grant
Method for verification of crosstalk noise in a CMOS design
Issue Date:2002-12-24
•12 Claims
•6 Drawing Sheets
Abstract
In deep submicron technologies, coupling capacitance significantly dominates the total parasitic capacitance. This causes crosstalk noise to be induced on quiescent signals which could lead to catastrophic failures. A methodology is provided that is a practical approach to full-chip crosstalk noise verification. A grouping based method is described for identification of potential victims and associated aggressors, using either timing information or functional information. Potential victim signal lines are selected and pruned based on total coupling capacitance to various signal groups.
Metadata
Assignee
- Texas Instruments Incorporated
Inventors
- Nagaraj N. Savithri
- Franciso A. Cano
Application Information
Application Number:US 09/608,309
Filing Date:2000-06-30
Priority Date:1999-07-15
Art Unit:7
Classifications
IPC:
G06F 1750
Field of Search:
716 5716 6716 8
Patent Drawings (6 sheets)
Description
[0002] This application claims priority under 35 USC §119(e)(1) of Provisional Application No. 60/143,972, filed Jul. 15, 1999 and Provisional Application No. 60/153,599, filed Sep. 13, 1999.
Cross Reference to Related Applications
[0003] This application is related to co-assigned U.S. Pat. No. 6,378,109 filed contemporaneously herewith and incorporated herein by reference.
Technical Field of the Invention
[0004] This invention relates to methods for designing and fabricating digital circuits, and in particular to simulation and analysis of the circuit design in order to detect and eliminate crosstalk induced by capacitive coupling.
Background of the Invention
[0005] Before an integrated circuit is fabricated, a trial design is first analyzed by simulating the operation of the proposed circuits that are to be included within the integrated circuit. Simulation is used to verify correct functional operation of the circuit, as well as to verify correct dynamic timing operation. When two signal lines on an integrated circuit are physically adjacent there is a capacitance between the signal lines that may cause signal interference due to signal coupling.
[0006] Signal integrity is one of the key challenges in design and test, now and beyond 100 nm technologies. Rapid technology scaling is causing increased coupling capacitances due to reduced signal to signal spacing and increased distance from the substrate. Among several types of noise, crosstalk noise introduced due to parasitic coupling is predominant in digital designs. Interconnect parasitic coupling has two effects, at least. A first effect is on the delay due to different switching patterns and this affects the performance of the chip besides potential race conditions. A second effect is induced glitches or noise that could cause functional failure by switching logic-states. In the case of designs employing dynamic logic design styles, this problem is more severe due to increased noise sensitivities of pre-charged nodes. In the case of static logic, except for asynchronous signals such as preset/clear and clock nodes, a failure could result only when data inputs connected to storage elements are latched during the active phase of the clock.
[0007] Design complexities are increasing with system on chip (SOC) designs and increased length of interconnect at full-chip level is one of the major sources of crosstalk noise problem. This refers to integration of system level functions, for example, many functions in a cellular telephone integrated into one integrated circuit (IC). Typically, SOC designs have CPU cores, memory blocks, logic blocks and possibly analog design blocks, all on one chip. Dynamic simulation of large designs with millions of parasitics is computationally prohibitive; in order reduce computational loads, static noise verification is used. Magnitude of the parasitic coupling, behavior of victim drivers during noise injection and dependency between factors like timing and parasitic coupling are some of key challenges in noise verification.
[0008] A commonly used simulator for designing integrated circuits is SPICE, which is available from the University of California at Berkeley, via the Department of Electrical Engineering and Computer Sciences. However, a SPICE simulation of all the nets in an entire chip is far too complex to be practical.
[0009] In order to perform a SPICE simulation of a circuit, all of the nodes between every component in the circuit need to be numbered. Then those nodes, the type of components at each node, and the component magnitudes are entered into the SPICE program.
[0010] If the circuit which is being simulated is an entire integrated circuit chip, then the number of nodes and corresponding components which need to be entered into the SPICE program is overwhelming. Firstly, the number of nets may be 40,000 in a current typical integrated circuit design. Secondly, for each such net, about seventy discreet components need to be entered because in the actual chip, the net components are distributed. Specifically, each signal line has capacitance which is distributed throughout the line; and each signal line also has a resistance which is distributed throughout the line. To simulate these distributed components, each signal line needs to be represented by a RC ladder circuit which has about two dozen nodes; with each node having a resistor to the next node, a capacitor to ground, and a capacitor to any adjacent signal line.
[0011] After all of the nodes and corresponding components for all the nets are entered into the SPICE program, the program operates to determine the voltages which occur on each node in sequential increments of time. Typically, about 1,000 increments of ten picoseconds each are needed to obtain the entire voltage waveform on a node in one net in an integrated circuit chip. To determine the voltages for just one time increment the SPICE program repetitively solves a matrix equation which is of the form [Y][V]=[1]. Here, Y is an n-×-n matrix, V is an n×1 matrix, and I is an n×1 matrix; where n is the number of nodes in the circuit. Thus, for a single victim net with twenty aggressor nets, n is (24 nodes per net)×(21 nets) or 504.
[0012] For each increment in time the SPICE program makes about five iterations before it converges on a solution. This iterative process is repeated for each of the subsequent time increments. Using a current state of the art workstation, it takes about ten minutes to perform a SPICE simulation of a single circuit which has 500 nodes and for which a solution is sought for 1,000 time increments. Such a circuit represents a typical victim net with twenty aggressor nets. Consequently, to simulate a chip which has 40,000 nets would take about 400,000 minutes, or more than 270 days to complete!
[0013] Accordingly, a primary object of the present invention is to provide a method of designing circuit chips by which the above problems are overcome.
[0014] Other objects and advantages will be apparent to those of ordinary skill in the art having reference to the following figures and specification.
Summary of the Invention
[0015] Pruning to identify potential victims and associated aggressors forms a key aspect of the methodology of the present invention. In general, and in form the present invention, a method of fabricating an integrated circuit is provided which contains a plurality of signal lines in close proximity, such that capacitive coupling among the signal lines is operable to induce crosstalk on at least one of the signal lines, wherein parasitics are extracted from a trial layout of the integrated circuit, the method comprising the steps of: grouping the plurality of signal lines into a plurality of aggressor groups; pruning the plurality of signal lines to form a plurality of victim signal lines; building a minimum region network for each victim signal line of the plurality of victim signal lines comprising the respective victim signal line, aggressor signal lines associated with the respective victim signal line, and associated parasitics; and simulating the operation of each minimum region network to determine an amount of noise induced on each respective victim signal line by the aggressor signal lines associated with the respective victim signal line.
[0016] In another form of the invention, an integrated circuit is fabricated according to a modified trial layout.
[0017] In another form of the invention, a computer system is provided which as a mass storage device that holds a design program for designing an integrated circuit according to the above described method.
[0018] These and other features of the invention that will be apparent to those skilled in the art from the following detailed description of the invention, taken together with the accompanying drawings.
Description of the Drawings
[0019] FIG. 1 is an illustration of a computer system which contains a design program for designing integrated circuits incorporating aspects of the present invention;
[0020] FIG. 2 is a block diagram of the computer of FIG. 1;
[0021] FIG. 3 illustrates a schmoo depicting a crosstalk induced failure;
[0022] FIG. 4A illustrates a conceptual block diagram of a crosstalk verification methodology, according to aspects of the present invention;
[0023] FIG. 4B is a process chart illustrating the methodology of FIG. 4A in more detail;
[0024] FIG. 5 is a coupled simulation model for determining propagation delay and noise induction in a signal line influenced by signal transitions on adjacent signal lines, according to an aspect of the present invention;
[0025] FIG. 6 illustrates a minimum-region network;
[0026] FIG. 7 is a process diagram illustrating the methodology of FIG. 4A and FIG. 4B in more detail;
[0027] FIG. 8 is a timing diagram illustrating switching events on three coupled signals;
[0028] FIGS. 9A and 9B are schematic diagrams of examples of grouping without timing information;
[0029] FIG. 10 shows a distribution of percentage coupling with respect to total coupling in an example complex integrated circuit; and
[0030] FIG. 11 is a block diagram of an integrated circuit that is designed and fabricated according to aspects of the present invention.
[0031] Corresponding numerals and symbols in the different figures and tables refer to corresponding parts unless otherwise indicated.
Detailed Description of the Invention
[0032] Aspects of the present invention include methods and apparatus for designing an integrated circuit. In the following description, specific information is set forth to provide a thorough understanding of the present invention. Well-known circuits and devices are included in block diagram form in order to not to complicate the description unnecessarily. Moreover, it will be apparent to one skilled in the art that specific details of these blocks are not required in order to practice the present invention. As discussed above, interconnect parasitic coupling has two effects, at least, that are important in the design of an integrated circuit. A first effect is on the delay due to different switching patterns and this affects the performance of the chip besides potential race conditions. The second effect is induced glitches or noise that could cause functional failure by switching logic-states. A method for improving simulation results in view of the second effect is described herein. Methods for improving simulation results in view of the first effect is disclosed in co-assigned U.S. Pat. No. 6,253,359 entitled Method for Analyzing Circuit Delays Caused by Capacitive Coupling in Digital Circuits and is incorporated herein by reference. Methods described in '993 may be advantageously combined with methods described herein.
[0033] Simplified models for crosstalk are described by Ashok Vittal and Malgorzata Marek-Sadowska in Crosstalk Reduction for VLSI,in IEEE Trans. on CAD, March 1997, for example. Reduced order modeling of parasitic networks is covered by Altan Odabasioglu, Mustafa Celik, and Lawrence T. Pileggi in PRIMA: Passive Reduced-order Interconnect Macromodeling Algorithm,in Proc. of ICCAD '97, for example. A method to account for circuit functionality during noise analysis is covered by D. A. Kirkpatrick and Alberto Vincentelli in Digital Sensitivity Predicting Signal Interaction Using Functional Analysis,in Proc. of ICCAD '96 A method of combining reduced order models and non-linear drivers is discussed by Lun Ye, Foong-Charn Chang, Peter Feldmann, Rakesh Chadha, N. S. Nagaraj in Chip-Level Verification of Parasitic Coupling Effects in Deep Sub-micron Digital Designs,in Proc. of DATE '99. Noise analysis based on noise propagation is described by Kenneth L. Shepard and Vinod Narayanan in Noise in Deep Submicron Digital Design,Tech. Dig. IEEE/ACM Int. Conf., Computer-Aided Design. November 1996. Methods to minimize crosstalk and crosstalk-based routing are covered by Joon-Seo Yim, Chong-Min Kyung in Reducing Cross-Coupling Among Interconnect Wires in Deep-Submicron Datapath Design,in Proc. of DAC 1999, for example. A practical approach to crosstalk noise verification has now been discovered which uses a pruning method. Its application to crosstalk noise verification will now be described in detail.
[0034] FIG. 1 is an illustration of a computer system 1000 which contains a design program incorporating aspects of the present invention, and FIG. 2 is a block diagram of the computer of FIG. 1. Computer system 1000 includes monitor 1004, system box 1006, keyboard 1008 and pointing device 1009. Within system box 1006 central possessing unit (CPU) 1140 with embedded resource 1143 communicates via bus 1145 to memory 1141 and to I/O bridge 1142. I/O bridge 1142 is connected to I/O bus 1146 and thereby to cursor control device 1150, keyboard interface 1151, hard disk drive 1153, compact disk drive 1153, display driver 1154 and network interface 1156. Network interface 1156 can be connected to an external interface via connection 1160. A design program that contains steps for designing an integrated circuit according to aspects of the present invention, as described in the following paragraphs, is stored on a hard drive 1152. This design program can be introduced into a computer 1000 via a compact disk installed in a compact disk drive 1153, or down loaded via network interface 1156, or by other means, such as a floppy disk or tape, for example. The program is transferred to memory 1141 and instructions which comprise the program are executed by processor 1140. Portions of the integrated circuit design are displayed on monitor 1004. The design program includes a simulator for modeling and extracting parasitic effects and simulating the operation of the integrated circuit according to aspects of the present invention.
[0035] Crosstalk noise is a transient event and is highly dependent on timing relationship between coupled signals. FIG. 3 shows a schmoo (a plot of frequency vs. voltage of operation) that depicts how crosstalk noise could manifest in practice in the operation of an integrated circuit (IC). Depending on the timing relationship, induced noise could cause an operational failure of the IC at certain frequencies and voltages, such as in region 300, whereas the IC could operate correctly at neighboring frequencies outside of region 300. This could be hard problem to detect during product qualification as a schmoo is generally created at discrete frequencies. Also, a complete set of test vectors used to exercise the IC may not be complete during product qualification, which could leave room for failure if crosstalk noise is not detected during the design phase itself.
[0036] Table 1 defines several terms that are used throughout this document.
| TABLE 1 |
|---|
| Definition of Terms |
[0037] The crosstalk verification methodology described herein emphasizes pruning the total number of nets in a proposed integrated circuit design to find potential victim clusters and to analyze these clusters independently. A victim cluster is a victim and its associated aggressors. A conceptual flow diagram for this methodology is described in FIG. 4A.
[0038] Step 400, Coupled Network Extraction, involves detailed parasitic extraction of the entire design including all the coupled parasitics. Typically, this is the most time consuming portion as some form of quasi-3-D or 2.5D approach is used.
[0039] Step 410, Victim/Aggressor Selection, involves identification of potential victims and associated aggressors. Both spatial and temporal correlations are typically used in this process. This reduces the problem space and helps focus on potential problems.
[0040] Step 420, Cluster Network Generation and Sensitization, involves building a “minimum region network” which constitutes the victim, aggressors and associated R(L)C parasitics. Minimum region networks could be as simple as only the channel connected components, i.e., the source/drain connected transistors that make up the output stage of a driver, or can include complete drivers for victims/aggressors as shown in FIG. 6.
[0041] Step 430, Crosstalk Noise Computation, involves actual crosstalk noise computation and checking for noise propagation. This is typically done using analytical models or some form of electrical simulation. Techniques for reduced order modeling of parasitic networks is known and are helpful in reducing computation times. Failure is typically determined based on noise margins of the victim receivers and using set-up time information from timing analysis.
[0042] FIG. 4B is a process chart illustrating the methodology of FIG. 4A in more detail. A trial layout of an integrated circuit design is completed in step 440, using known layout tools. In step 400, a parasitic extraction step 401 produces a parasitic database 402 that is translated for use by other simulation modules by translation step 403. In step 410, filter step 411 makes use of logic correlation step 412 and temporal correlation step 413 to identify potential victim nets and associated aggressor nets. In step 420, a minimum region network is built in step 421 using command file information 422, and library information 423. Since a minimum region net is truncated from the rest of the circuitry, tie-off schemes and stimulus models are provided in step 424. In step 430, various analytical models, such as Spice 431a,or other reduced order modeling tools 431nare used to simulate crosstalk noise and to check for noise propagation. In step 450, victim nets which are susceptible to crosstalk noise are identified based on the results of step 430.
[0043] FIG. 5 is a coupled simulation model 200 for determining propagation delay and noise induction in a signal line influenced by signal transitions on adjacent signal lines using parasitic values extracted in extraction step 400, according to an aspect of the present invention. Simulation model 200 is a fully distributed model so that effects of signal coupling can be correctly modeled. Signal line 203 which is referred to as a “victim net” is modeled as a set of capacitors 205a˜n to ground and a set of resistors 207a-nin series. Victim net 203 includes buffer 201 connected to receive input signal 208 and inverter 202 connected via signal 209 to an output load represented as capacitor Cload. Signal line 213 and signal line 223 represent signal lines routed adjacent to victim net 203. These are referred to as “aggressor nets.” Both aggressor nets are also modeled as a set of distributed capacitors 214a-214n,224a-224nand resistors 217a-217n,227a-227n.However, the values of distributed capacitors 214a-nare scaled to account for coupling capacitance between aggressor net 213 and another signal line adjacent to it. Likewise, the values of distributed capacitors 224a-nare scaled to account for coupling capacitance between aggressor net 223 and another signal line adjacent to it. Aggressor net 213 includes buffer 211 connected to receive input signal 218 and inverter 212 connected via signal 219 to an output load represented as capacitor Cload. Victim net 223 includes buffer 221 connected to receive input signal 228 and inverter 222 connected via signal 229 to an output load represented as capacitor Cload.
[0044] Still referring to FIG. 5, coupling capacitance between victim net 203 and aggressor net 213 is modeled as distributed capacitors 206a-n.Likewise, coupling capacitance between victim nets 203 and aggressor nets 223 is modeled as distributed capacitors 205a-n.
[0045] Still referring to FIG. 5, a series of SPICE simulations can be preformed in which the time at which input signals to the aggressor nets switch on lines 218 and 228 is varied with respect to the time at which an input signal to the victim net on line 208 switches to determine a worst case propagation delay 240. However, with current computer systems, it is impractical to perform such a series of SPICE simulations, or other similar types of simulators, using distributed models on every possible victim net in an integrated circuit.
[0046] FIG. 6 is an illustration of a minimum-region network. As discussed above, a minimum region network can be as simple as the channel connected components or can include complete drivers for victims/aggressors. Minimum region network 500 includes victim signal 510 and victim driver 511, along with aggressor signals Aggr-1, Aggr-2 and Aggr-3 and corresponding aggressor drivers 521, 531, and 541. A minimum region network is typically reduced to smaller networks for increased computing efficiency in step 406 by truncating victim receiver 512 and aggressor receivers 522, 532, and 542; however, victim receiver 512 could also be included to verify propagation of noise for accurate delay. A Driving Point Admittance (DPA or pi-model) is sufficient to be used for victim receiver output load to analyze noise propagation.
[0047] One of the major contributors to faster signal integrity verification at gate level is use of a simplified model for the driver instead of transistor level description. However, a resulting error in simulated crosstalk noise could be significant as the victim driver is essentially not making a rail-to-rail transition. Error due to driver modeling can be easily quantified using a simple controlled evaluation model, similar to FIG. 5. In this evaluation model, parasitics such as resistance, coupling capacitance and capacitance to ground are parameterized to enable evaluation for different technology nodes. By using different drive strengths and different types of cells for aggressors and victim drivers, it can be easily seen that using linearized models for drivers can result in significant errors as high as 50% in the region of crosstalk failure. A method to obtain simplified non-linear drivers and reduction in error is described in “Chip-Level Verification of Parasitic Coupling Effects in Deep Sub-micron Digital Designs,” was noted earlier and is incorporated herein by reference.
[0048] FIG. 7 is a process diagram illustrating the methodology of FIG. 4A and FIG. 4B in more detail. A trial layout of an integrated circuit design is completed in step 740, using known layout tools. A parasitic extraction step 701 produces a parasitic database 702. Various tools for parasitic extraction are known to those skilled in the art, such as “maskPE”. Parasitic extraction step 701 is based on empirical equations for parasitic capacitance. Interconnect process parameter rules 704 for metal and dielectrics that yield maximum capacitance within specified parameter limits of the selected process are used for extraction. A flat extraction method is used to account for all the interconnect interactions in various levels of interconnect hierarchy. On complex IC designs, such as digital signal processor (DSP) core, a complete coupled parasitic extraction is performed. In order to address capacity issues for full-chip verification, a two-pass parasitic extraction is used. The first pass performed in step 766 extracts coupled line-to-line parasitic capacitances, which includes only horizontal components. A second pass performed in step 701 extracts uncoupled parasitic capacitance and resistance, which includes both horizontal and vertical components. This provides 10-20× reduction in number of devices extracted as compared to a complete coupled RCL extraction. This is motivated by the fact that in deep sub-micron (DSM) technologies, metal lines are becoming taller and thinner which increases the lateral component of the capacitance. Typically, designs implemented in technologies below 0.5 micron are referred to as DSM designs. Currently, 0.18u and 0.13u are some of the leading edge technologies. Crosstalk becomes a more significant issue in DSM designs. A tabulation of capacitance data, such as Cg (Capacitance to ground), Cc (Coupling capacitance), Ctot (Total capacitance), and percentage coupling to total capacitance (PCCT) is determined for every signal and saved in capacitance database 721.
[0049] FIG. 7 is a process diagram illustrating he methodology of FIG. 4A and FIG. 4B in more detail. A trial layout of an integrated circuit design is completed in step 740, using known layout tools. A parasitic extraction step 701 produces a parasitic database 702. Various tools for parasitic extraction are known to those skilled in the art, such as “maskPE”. Parasitic extraction step 701 is based on empirical equations for parasitic capacitance. Interconnect process parameter rules 704 for metal and dielectrics that yield maximum capacitance within specified parameter limits of the selected process are used for extraction. A flat extraction method is used to account for all the interconnect interactions in various levels of interconnect hierarchy. On complex IC designs, such as digital signal processor (DSP) core, a complete coupled parasitic extraction is performed. In order to address capacity issues for full-chip verification, a two-pass parasitic extraction is used. The first pass performed in step 766 using rules 765 extracts coupled line-to-line parasitic capacitances, which includes only horizontal components, which are saved in horizontal coupling capacitance database 760. A second pass performed in step 701 extracts uncoupled parasitic capacitance and resistance, which includes both horizontal and vertical components. This provides 10-20× reduction in number of devices extracted as compared to a complete coupled RCL extraction. This is motivated by the fact that in deep sub-micron (DSM) technologies, metal lines are becoming taller and thinner which increases the lateral component of the capacitance. Typically, designs implemented in technologies below 0.5 micron are referred to as DSM designs. Currently, 0.18u and 0.13u are some of the leading edge technologies. Crosstalk becomes a more significant issue in DSM designs. A tabulation of capacitance data, such as Cg (Capacitance to ground), Cc (Coupling capacitance), Ctot (Total capacitance), and percentage coupling to total capacitance (PCCT) is determined for every signal and saved in capacitance database 721.
[0050] Selection of potential victims and their associated aggressors is a crucial step in the crosstalk verification methodology according to an aspect of the present invention, which is performed in FindVictims filtering step 711. Pruning efficiency is extremely important to reduce the crosstalk noise computation time, while not missing victims. A concept of grouping is used to perform victim/aggressor selection. A group is defined as a set of signals that could switch at the same time and hence collectively induce a glitch on a victim that is larger than if the aggressors switched at dispersed times. Timing information 712 can be obtained from commercial timing analyzers such as Pearl (Cadence) and PrimeTime (Synopsys), as represented by 761, 762.
[0051] As an example, refer to FIG. 8, which is a timing diagram illustrating switching events 800, 810 and 820 on three coupled signals A, B, and C, respectively. Each switching event includes a range of signal transitions representative of high to low, low to high, best case and worst case timing. In FIG. 8 for glitch analysis, if a victim signal is latched by down-stream circuitry in response to a clock edge shortly after it stabilizes, then it must be stable for a specified set-up and hold-time before and after the clock edge, respectively. Coupled aggressor signals which have switching events during the set-up and hold-time period of a victim signal may induce noise on the victim during this critical period which can cause a failure, thus such a signal would be classified as an aggressor signal. Conversely, switching events which occur on a coupled signal before the victim signal has switched and stabilized would not cause a failure and are therefore not classified as aggressor signals. Thus, for signal A, only B can be an aggressor, for signal B, neither A or C can be aggressors and for signal C, both A and B are aggressors. For a given victim, all the aggressors that switch after the victim has settled are grouped. All signals for which timing information does not exist are grouped as an asynchronous (ASYNC) group and are always treated as aggressors. Ultimately, a victim window may be selected as all time greater that the earliest required setup time to all down stream latches.
[0052] Referring again to FIG. 8 for delay calculation, a timing window can also be associated with a potential victim signal to allow pruning of potential aggressor signals which do not have transitions within the window of a victim. For example, window 840 has a width W and is associated with signal C. If a potential aggressor signal does not transition within the window, it is excluded from delay analysis. In this case, signal A would be classified as an aggressor to C, but signal B would be excluded from being an aggressor to C.
[0053] Referring again to FIG. 8, if a victim signal is not latched by clock signal, but is instead received by dynamic circuitry, for example, then timing information is evaluated in a different manner, since noise induced crosstalk could trigger a dynamic circuit anytime after the circuit has been pre-charged. Therefore, in FIG. 8, for signal A, signals B or C could be classified as aggressors in a dynamic circuit. Likewise for signal B, signals A and C could be classified as aggressors, depending on pre-charge timing.
[0054] FIGS. 9A and 9B are schematic diagrams of examples of grouping without timing information. Grouping can also be used when timing information is not available. For example, referring to FIG. 9A, signals 900-903 from a set of flip-flops 910-913 which are clocked by the same clock signal CLK are considered as a group, since all the flip-flop outputs could potentially switch at the same time. Similarly, referring to FIG. 9B, a set of signals in a bus 920 is considered as a group since bits of the same bus could potentially switch at the same time; this is referred to as name based grouping, or function based grouping. As shown in FIG. 9B, a signal 930 crossed over by bus 920 could be a potential victim even though crossover coupling capacitance to each individual bus signal is small.
[0055] As another example, all the signals that are strictly static, such as those tied to tie-off cells, are grouped to signify that they can never be aggressors. Many different types of groups can be identified based on operating characteristics of a given IC. One skilled in the art may identify additional groups which could have significant crosstalk impact on coupled victim signals in accordance with aspects of the present invention, given the preceding examples.
[0056] Referring again to FIG. 7, groups of signals are identified in BuildGroup step 715 using netlist information 716, as described above. Netlist information 716 includes a list of all of the nets in a given IC design and their connectivity to circuits in the design. Each group is categorized with relevant group information in step 717. Once the groups of signals are identified, potential victims and their associated aggressors are determined in filter step 711 as follows. For a potential victim signal, if Ccis the total coupling capacitance to a potential aggressor signal or to a potential aggressor group, Cgis the total capacitance to ground and Cpis the total pin capacitance of the receiver input pins, then total capacitance for that potential victim signal Ctis Cc+Cg+Cp. For N potential aggressor signals or potential aggressor groups, a given signal is a victim if its cumulative percentage coupling with respect to total coupling (PCCT) is greater than a selected group threshold; that is, if
[0057] where pgis a group percentage threshold value for PCCT. To control the number of individual aggressors that have very small percentage coupling on their own and thereby reduce the computation burden, a signal percentage threshold psis selected (default value 0). For a given victim, an individual signal or group of signals from among the N groups of signals is an aggressor if its PCCT is greater than ps; that is, if
[0058] Threshold values are selected in threshold step 713. It has been determined that using a value of 1% for psprovides good results. For a given IC design, threshold values may be selected based on the type of circuit a potential victim signal is associated with, such as a dynamic circuit or a clocked circuit. Thresholds may be raised or lowered, based on the noise immunity of the circuit technology.
[0059] When timing information is available, the number of groups can be considered as two, where the first group satisfies timing requirements to be aggressors and the second group contains asynchronous signals.
[0060] In order to determine total coupling capacitance between a potential victim signal and each group of potential aggressor signals, a matrix of signal to signal coupling can be formed, based on capacitance database 721. Alternatively, instead of building matrix of signal to signal coupling (which could be really huge, e.g., 1010combinations for 100 K signals), capacitance data 721 can be streamed out on a per signal basis such that all C parasitics associated with a signal are listed exclusively in one delineated group.
[0061] Additional filtering can be performed in filter step 711. For example, drive strengths can be incorporated as a filtering step, with appropriate thresholds selected in threshold step 713. However, a high degree of non-linearity in crosstalk waveforms complicates the use of drive; strength thresholds. Another threshold based on total Cccan be selected in threshold step 713 and used to eliminate short signals. Advantageously, cumulative percentage coupling, pg, provides a conservative metric for crosstalk noise. Filtering based on spatial proximity can also be performed, for example.
[0062] Still referring to FIG. 7, after all of the victim signals are identified in filter step 711, victim clusters are defined in step 720, as described previously with respect to building a minimum region network. For each victim, an RC network similar to FIG. 5 in the form of a coupled RC network is generated by a parameter generation (paragen) step 722. Paragen 722 reads parasitic database 702 and then outputs a set of coupled RC networks 723. Alternatively, paragen 722 can output the minimum region networks as a SPEF dataset 724, where SPEF is Standard Parasitic Exchange Format, an IEEE standard format for representing parasitics.
[0063] Due to the accuracy issues involved in modeling of drivers, SPICE is used as the basic method to compute crosstalk noise in step 730. Sensitization vectors generated from library characterization data 731 is used to sensitize the aggressors and victims. Essentially, this data specifies how the input pins of cells in a minimum network should be tied to create a known output state. SPICE models for cells such as inverters, gates, etc, are obtained from sub-circuit database 732. An assumption that all the aggressors switch at the same time and in the same direction results in worst-case crosstalk. This is the normal mode of operation that enables worst-case analysis. In an embodiment of the present invention, all the inputs of a cluster are made to switch at the same time and the time difference between the outputs are measured. If the time difference between outputs are significantly large, then input arrival times can be changed to align all output transitions. Advantageously, switching windows can be used to reduce pessimism.
[0064] Alternatively, a faster crosstalk analysis solution based on MPVL, a commercially available Signal Integrity Verification (SIV) tool, is performed in step 725 using non-linear driver models 726 for aggressors and receivers. This can be used as an additional filter for crosstalk computation. Benchmarks of MPVL solution on just an RC network has shown a maximum error of 1% as compared to SPICE.
[0065] Networks having crosstalk noise violations are tabulated in step 750. In response to a detected crosstalk noise violation, the design of the IC can be modified to eliminate the violation by changing wire spacing or by insertion of repeaters in the victim signal, for example.
[0066] As discussed earlier, in order to minimize computation time, a two-pass extraction methodology is used for complicated ICs. The first-pass extraction includes only extraction of line-to-line couplings and total capacitance from lumped-RC extraction and is used to perform pruning. The second pass extraction includes all the selected victims and aggressors and is used for crosstalk noise analysis. FIG. 10 shows a distribution of percentage coupling with respect to total coupling (PCCT) for an IC design which includes a DSP, memory circuits, and other logic circuits. In this design 3.17% of signals were chosen as potential victims from the pruning process from the first-pass extraction based on a percentage coupling threshold of 50%. 10.11% of signals constituting victims and their associated aggressors were re-extracted in the second-pass extraction after the formation of minimum regions networks associated with the selected victims. The process described in FIG. 7 was repeated a number of times in order to eliminate crosstalk violations. After determining failed victims in step 750, the IC design was modified and the process repeated. The number of violations found during each compile (check-point in design stage) is shown in Table 2. An increase in the number of violations in the second compile is attributed to functionality changes. Victim violations were either signed-off after careful examination of timing relationships or corrected by increasing spacing or insertion of repeaters.
| TABLE 2 |
|---|
| Crosstalk Noise Violations on a DSP Chip |
[0067] Timing-based pruning is very important to reduce pessimism and thereby reduce the total amount of SPICE crosstalk simulation that must be done. For example, in another trial, pruning on a DSP chip with 88.3% of signals having timing information, the number of potential victims was reduced by 23% as compared to spatial grouping only. On a small random-logic block, timing-based pruning reduced number of potential victims from 111 to 7. Typically, a 10-25% reduction has been observed on large designs.
[0068] In another experiment, an analysis of the impact of intra-layer dielectric on crosstalk noise was performed using the methodology described herein. Analysis showed that a 27% change in dielectric constant changed the number of potential victims by about 37% for the same pruning criteria. Average crosstalk changed by 23% and the largest value of crosstalk changed by 14%.
[0069] In general, higher noise margins inherent to static CMOS designs helps minimize the impact on the design. However, in addition to parasitic coupling, increased resistance on long victim signals results in higher crosstalk noise. A well tuned design with good transition times on signals helps minimize noise problems. Noise on asynchronous signals such as reset/clear and clocked nodes need rigorous analysis with reduced noise thresholds.
[0070] Pruning to identify potential victims plays a crucial role in not only minimizing the scope of the problem, but also ensure that all the realistic aggressors are accounted for. The grouping approach to pruning has been very effective in the absence of timing information and for manual inspection of most sensitive victims. Timing-based pruning is critical for reducing pessimism, especially on random logic modules. Crosstalk analysis on independent clusters helps to parallelize computation. Noise propagation could be achieved by instantiating the receivers and DPA model. As most of the reduced order modeling techniques provide reasonably accurate interconnect models driver models contribute significantly to overall accuracy of crosstalk noise.
[0071] It has been found that number of potential victims in a large SOC type of designs, is typically 3-8% of total number of signals, of which typically less than 1% of signals turn out to be real problems. Proper design guidelines for spacing/repeaters and signal transition times often helps minimize the number of potential victims and resulting failures.
[0072] FIG. 11 is a block diagram of a microprocessor 1, which can be designed using aspects of the present invention. Microprocessor 1 is a VLIW digital signal processor (“DSP”). In the interest of clarity, FIG. 1 only shows those portions of microprocessor 1 that are relevant to an understanding of an embodiment of the present invention. Details of general construction for DSPs are well known, and may be found readily elsewhere. For example, U.S. Pat. No. 5,072,418 issued to Frederick Boutaud, et al, describes a DSP in detail and is incorporated herein by reference. U.S. Pat. No. 5,329,471 issued to Gary Swoboda, et al, describes in detail how to test and emulate a DSP and is incorporated herein by reference. Details of portions of microprocessor 1 relevant to an embodiment of the present invention are explained in sufficient detail hereinbelow, so as to enable one of ordinary skill in the microprocessor art to make and use the invention.
[0073] In microprocessor 1 there are shown a central processing unit (CPU) 10, data memory 22, program memory 23, peripherals 60 and an external memory interface (EMIF) with a direct memory access (DMA) 61. CPU 10 further has an instruction fetch/decode unit 10a-c,a plurality of execution units, including an arithmetic and load/store unit D1, a multiplier M1, an ALU/shifter unit S1, an arithmetic logic unit (“ALU”) L1, a shared multiport register file 20afrom which data are read and to which data are written. Decoded instructions are provided from the instruction fetch/decode unit 10a-cto the functional units D1, M1, S1, and L1 over various sets of control lines which are not shown. Data are provided to/from the register file 20afrom/to to load/store units D1 over a first set of busses 32a,to multiplier M1 over a second set of busses 34a,to ALU/shifter unit S1 over a third set of busses 36aand to ALU L1 over a fourth set of busses 38a.Data are provided to/from the memory 22 from/to the load/store units D1 via a fifth set of busses 40a.Note that the entire data path described above is duplicated with register file 20band execution units D2, M2, S2, and L2. Instructions are fetched by fetch unit 10afrom instruction memory 23 over a set of busses 41. Emulation unit 50 provides access to the internal operation of integrated circuit 42 which can be controlled by an external test system 51.
[0074] Note that the memory 22 and memory 23 are shown in FIG. 11 to be a part of a microprocessor 1 integrated circuit, the extent of which is represented by the box 42. The memories 22-23 could just as well be external to the microprocessor 1 integrated circuit 42, or part of it could reside on the integrated circuit 42 and part of it be external to the integrated circuit 42. Also, an alternate number of execution units can be used.
[0075] When microprocessor 1 is incorporated in a data processing system, additional memory or peripherals may be connected to microprocessor 1, as illustrated in FIG. 11. For example, Random Access Memory (RAM) 70, a Read Only Memory (ROM) 71 and a Disk 72 are shown connected via an external bus 73. Bus 73 is connected to the External Memory Interface (EMIF) which is part of functional block 61 within integrated circuit 42. A Direct Memory Access (DMA) controller is also included within block 61. The DMA controller is generally used to move data between memory and peripherals within microprocessor 1 and memory and peripherals which are external to microprocessor 1.
[0076] With technology scaling, crosstalk noise has become a problem in a DSM design such as integrated circuit 42. Unlike timing problems, crosstalk noise could manifest only intermittently when certain combinations of signal switching are satisfied. Such failures could happen in the end-application as testers fail to detect them during product qualification. An aspect of the present invention advantageously provides a practical method to perform static noise verification on integrated circuit 42 before it is fabricated. Advantageously, pruning methods based on grouping enables better coverage of potential victims and their associated aggressors, even in absence of timing. For example, buses such as 32a,34a,36a38a,40aand 41 can be grouped for victim pruning. Advantageously, timing-based pruning generally reduces pessimism around 30%; required timing information can be determined for integrated circuit 42 using simulation.
[0077] Several example systems which can benefit from aspects of the present invention are described in U.S. Pat. No. 5,072,418, which was incorporated by reference herein, particularly with reference to FIGS. 2-18 of U.S. Pat. No. 5,072,418. A microprocessor incorporating an aspect of the present invention to improve performance or reduce cost can be used to further improve the systems described in U.S. Pat. No. 5,072,418. Such systems include, but are not limited to, industrial process controls, automotive vehicle systems, motor controls, robotic control systems, satellite telecommunication systems, echo canceling systems, modems, video imaging systems, speech recognition systems, vocoder-modem systems with encryption, and such.
[0078] Fabrication of data processing device 1 after the design is completed and crosstalk violations have been dealt with according to aspects of the present invention, involves multiple steps of implanting various amounts of impurities into a semiconductor substrate and diffusing the impurities to selected depths within the substrate to form transistor devices. Masks are formed to control the placement of the impurities. Multiple layers of conductive material and insulative material are deposited and etched to interconnect the various devices. These steps are performed in a clean room environment.
[0079] A significant portion of the cost of producing the data processing device involves testing. While in wafer form, individual devices are biased to an operational state and probe tested for basic operational functionality. The wafer is then separated into individual dice which may be sold as bare die or packaged. After packaging, finished parts are biased into an operational state and tested for operational functionality.
[0080] Thus, a novel technique for pruning a large set of coupled signals to identify potential victims and their associated aggressors has been presented. This pruning process helps reduce the problem size and hence makes the crosstalk analysis faster. User-defined groups of signals are treated as aggressors during the pruning process. This is a very important and differentiating feature in this pruning process, as it helps in accounting for simultaneous switching of grouped signals. Bus signals can be grouped if no user-defined groups are specified. Usage of groups instead of individual signals helps in selection of more potential aggressors. This in turn helps in uncovering potential failures due to crosstalk.
[0081] Crosstalk verification is a complex problem involving several sources of error. It is important to know these sources of error and bound the crosstalk problem according to ensure circuit functionality. Sources of error include:
[0082] error in parasitic resistance and coupling capacitance
[0083] error in victim selection
[0084] error in obtaining worst case sensitization of drivers
[0085] error in obtaining switching windows for timing analysis
[0086] error in obtaining logic correlation between signals
[0087] error due to non-linear loads
[0088] error in SPICE models
[0089] error in selecting combinations of strong/weak/nominal SPICE models
[0090] process variation on interconnect and transistors.
[0091] Performing a worst case analysis in view of the above would result in highly pessimistic results and needlessly burden the design of an integrated circuit. Advantageously, in a crosstalk verification methodology embodying aspects of the present invention, thresholds are provided and can be selected to account for sources of error.
[0092] Advantageously, the number of aggressor groups that could act on a given victim can be controlled in the pruning process, in an embodiment of the present invention. Groups are generated based on the switching or transition patterns. All signals known to switch at the same time are grouped. Grouping of signals can be determined by net name and driver connectivity. However, generation of grouping information based on switching windows or transition information from other sources such as the following could also be done: static timing analysis, timing simulation, and functional simulation
[0093] For efficient processing of the large amounts of parasitic data (typically in millions of parasitic devices), every signal in the design is analyzed separately. This helps in better data management and faster processing. Some signals or groups may not be transitioning during the time when crosstalk on victim is catastrophic. Advantageously, these groups can be excluded during the pruning process.
[0094] Using groups during the pruning process could result in large number of individual signals for the R(L)C cluster network generation and detailed analysis. In order to overcome this problem, a signal threshold is used to eliminate signal which have little or no effect on a given victim signal.
[0095] The pruning step generates clusters of dominantly coupled signals for each potential victim. These clusters are used to construct a coupled R(L)C network and perform network analysis either using SPICE or other numerical methods. Advantageously, each cluster can be analyzed separately from the rest, and a set of cluster can be analyzed in parallel to reduce computation real time.
[0096] An assumption that all the aggressors switch at the same time and in the same direction results in worst-case crosstalk. This is the normal mode of operation that enables worst-case analysis. In an embodiment of the present invention, all the inputs of a cluster are made to switch at the same time and the time difference between the outputs are measured. If the time difference between outputs are significantly large, then input arrival times can be changed to align all output transitions. Advantageously, switching windows can be used to reduce pessimism.
[0097] As used herein, the terms “applied,” “connected,” and “connection” mean electrically connected, including where additional elements may be in the electrical connection path.
[0098] While the invention has been described with reference to illustrative embodiments, this description is not intended to be construed in a limiting sense. Various other embodiments of the invention will be apparent to persons skilled in the art upon reference to this description. It is therefore contemplated that the appended claims will cover any such modifications of the embodiments as fall within the true scope and spirit of the invention.
Claims
What is claimed is:
1. A method for designing an integrated circuit which contains a plurality of signal lines in close proximity, such that capacitive coupling among the signal lines is operable to induce crosstalk on at least one of the signal lines, wherein parasitics are extracted from a trial layout of the integrated circuit, the method comprising the steps of;
grouping the plurality of signal lines into a plurality of aggressor groups;
pruning the plurality of signal lines to form a plurality of victim signal lines, wherein the step of pruning comprises calculating a cumulative percentage coupling to total capacitance (PCCT) for each potential victim signal line from the plurality of signal lines relative to the plurality of aggressor groups; and classifying each potential victim signal line as a victim signal line only if the associated cumulative PCCT exceeds a selected threshold value;
building a minimum region network for each victim signal line of the plurality of victim signal lines comprising the respective victim signal line, aggressor signal lines associated with the respective victim signal line, and associated parasitics; and
simulating the operation of each minimum region network to determine an amount of noise induced on each respective victim signal line by the aggressor signal lines associated with the respective victim signal line.
2. The method of claim 1, further comprising the step of modifying the trial layout if the amount of noise induced on any of the victim signal lines exceeds a selected value.
3. The method of claim 2, further comprising the step of fabricating the integrated circuit according to the modified trial layout.
4. The method of claim 1, wherein the step of grouping comprises selecting as a first group a set of signal lines which form a bus in the integrated circuit.
5. The method of claim 1, wherein the step of grouping comprises selecting as a first group a set of signal lines which are outputs of a set of latches that are clocked by a same clock signal.
6. The method of claim 1, wherein the step of grouping comprises selecting as a first group a set of signal lines which have signal transitions occurring within a selected time window.
7. A method for designing an integrated circuit which contains a plurality of signal lines in close proximity, such that capacitive coupling among the signal lines is operable to induce crosstalk on at least one of the signal lines, wherein parasitics are extracted from a trial layout of the integrated circuit, the method comprising the steps of;
grouping the plurality of signal lines into a plurality of aggressor groups, wherein the step of grouping comprises not including a first signal line within an aggressor group for a first potential victim signal line if the first signal line has a percentage coupling to total capacitance (PCCT) relative to the first potential victim signal line that is less than a selected threshold value;
pruning the plurality of signal lines to form a plurality of victim signal lines, building a minimum region network for each victim signal line of the plurality of victim signal lines comprising the respective victim signal line, aggressor signal lines associated with the respective victim signal line, and associated parasitics; and
simulating the operation of each minimum region network to determine an amount of noise induced on each respective victim signal line by the aggressor signal lines associated with the respective victim signal line.
8. A computer system, comprising:
a processor for processing instructions;
a memory circuit for holding instructions connected to the processor;
a mass storage device for holding a design program operable to transfer the design program to the memory circuit;
wherein the design program on the mass storage device comprises instructions for a method of analyzing an integrated circuit design which contains a plurality of signal lines in close proximity, such that capacitive coupling among the signal lines is operable to induce crosstalk on at least one of the signal lines, wherein parasitics are extracted from a trial layout of the integrated circuit, the method comprising the steps of:
grouping the plurality of signal lines into a plurality of aggressor groups;
pruning the plurality of signal lines to form a plurality of victim signal lines, wherein the step of pruning comprises calculating a cumulative percentage coupling to total capacitance (PCCT) for each potential victim signal line from the plurality of signal lines relative to the plurality of aggressor groups; and classifying each potential victim signal line as a victim signal line only if the associated cumulative PCCT exceeds a selected threshold value;
building a minimum region network for each victim signal line of the plurality of victim signal lines comprising the respective victim signal line, aggressor signal lines associated with the respective victim signal line, and associated parasitics; and
simulating the operation of each minimum region network to determine an amount of noise induced on each respective victim signal line by the aggressor signal lines associated with the respective victim signal line.
9. The computer system of claim 8, wherein the step of grouping comprises not including a first signal line within an aggressor group for a first potential victim signal line if the first signal line has a percentage coupling to total capacitance (PCCT) relative to the first potential victim signal line that is less than a selected threshold value.
10. A mass storage device for holding a computer program, wherein the design program on the mass storage device comprises instructions for a method of analyzing an integrated circuit design which contains a plurality of signal lines in close proximity, such that capacitive coupling among the signal lines is operable to induce crosstalk on at least one of the signal lines, wherein parasitics are extracted from a trial layout of the integrated circuit, the method comprising the steps of:
grouping the plurality of signal lines into a plurality of aggressor groups;
pruning the plurality of signal lines to form a plurality of victim signal lines, wherein the step of pruning comprises calculating a cumulative percentage coupling to total capacitance (PCCT) for each potential victim signal line from the plurality of signal lines relative to the plurality of aggressor groups; and classifying each potential victim signal line as a victim signal line only if the associated cumulative PCCT exceeds a selected threshold value;
building a minimum region network for each victim signal line of the plurality of victim signal lines comprising the respective victim signal line, aggressor signal lines associated with the respective victim signal line, and associated parasitics; and
simulating the operation of each minimum region network to determine an amount of noise induced on each respective victim signal line by the aggressor signal lines associated with the respective victim signal line.
11. The mass storage device of claim 10, wherein the mass storage device is a diskette.
12. The mass storage device of claim 10, wherein the step of grouping comprises not including a first signal line within an aggressor group for a first potential victim signal line if the first signal line has a percentage coupling to total capacitance (PCCT) relative to the first potential victim signal line that is less than a selected threshold value.
Patent Citations (4)
Non-Patent Literature (14)
- Kenneth L. Shepard and Vinod Narayanan, Noise in Deep Submicron Digital Design, Tech Dig. IEEE/ACM Int. Conf., Computer-Aided-Design, Nov. 1996, pp. 524-531.Applicant
- Altan Odabasioglu, Mustafa Celik and Lawrence T. Pileggi, PRIMA: Passive Reduced-Order Interconnect Macromodeling Algorithm, in Proc. of ICCAD '97, pp. 58-65.Applicant
- P. Feldman and R.W. Freund, Reduced-Order Modeling of Large Linear Subcircuits Via a Block Lanczos Algorithm, IEEE/ACM Proc. DAC, pp. 474-479, Jun., 1995.Applicant
- I.M. Elfadel and D.D. Ling, A Block Rational Arnoldi Algorithm for Multipoint Passive Model-Order Reduction of Multiport RLC Networks, Proc. ICCAD, 1997, pp. 66-71.Applicant
- D.A. Kirpatrick and Alberto Vincentelli, Digital Sensitivity Predicting Signal Interaction Using Functional Analysis in Proc. of ICCAD '96, pp. 536-541.Applicant
- Lun Ye, Foong-Charn Chang, Peter Feldmann, Rakesh Chadha, N.S. Nagaraj, Chip-Level Verification of Parasitic Coupling Effects in Deep Sub-micron Digital Designs, in Proc. of Date '99, pp. 658-663.Applicant
- Hui-Ru Jiang, Jing-Yang Jou, Yao-Wen Chang, Noise-Constrained Performance Optimization by Simultaneous Gate and Wire Sizing Based on Lagrangian Relaxation, in Proc. of DAC, 1999, pp. 90-95.Applicant
- Prashant Saxena, C.L. Liu, Crosstalk Minimization Using Wire Perturbations, in Proc. Of DAC, '99, pp. 100-103.Applicant
- Joon-Seo Yim, Chong-Min Kyung, Reducing Cross-Coupling Among Interconnect Wires in Deep-Submicron Datapath Design, in Proc. of DAC '99, pp. 485-490.Applicant
- T. Sakurai, Closed Form Expressions for Interconnect Delay, Coupling and Crosstalk in VLSIs, IEEE Trans. Electron Devices, vol. 40, 1193, pp. 118-123.Applicant
- Ashok Vittal and Malgorzata Marek-Sadowska, Crosstalk Reduction for VLSI, in IEEE Trans. on CAD, Mar. 1997, pp. 290-298.Applicant
- Joardar, Kuntal, A Simple Approach to Modeling Cross-Talk in Integrated Circuits, IEEE Journal of Sold-State Circuits, vol. 29, No. 10, Oct. 1994, p. 1212+.Applicant
- van Genderen, A.J., et al.; Reduced RC Models for IC Interconnections With Coupling Capacitances, IEEE 3/92, p. 132-136 (0-8186-2645-392).Applicant
- The National Roadmap for Semiconductors (NTRS): Modeling and Simulation Chapter, 1997 Edition, Online Document (http://notes.sematech.org/97melec.htm), pp. 187-196.Applicant