US20090032903A1 - Multiple voltage integrated circuit and design method therefor - Google Patents

Abstract

An integrated circuit (IC) design, method and program product for reducing IC design power consumption. The IC is organized in circuit rows. Circuit rows may include a low voltage island powered by a low voltage (V_ddl) supply and a high voltage island powered by a high voltage (V_ddh) supply. Circuit elements including cells, latches and macros are placed with high or low voltage islands to minimize IC power while maintaining overall performance. Level converters may be placed with high voltage circuit elements.

Description

CROSS REFERENCE TO RELATED APPLICATION
• The present application is a continuation of allowed U.S. patent application Ser. No. 11/460,537, (Attorney docket No. YOR920030358US2) entitled “MULTIPLE VOLTAGE INTEGRATED CIRCUIT AND DESIGN METHOD THEREFOR” to Anthony Correale, Jr. et al., and a divisional application of U.S. Pat. No. 7,111,266, (Attorney docket No. YOR920030358US1) entitled “MULTIPLE VOLTAGE INTEGRATED CIRCUIT AND DESIGN METHOD THEREFOR” to Anthony Correale, Jr. et al.; and related to U.S. Pat. No. 7,089,510 (Attorney Docket No. YOR920030359US1) entitled “METHOD AND PROGRAM PRODUCT OF LEVEL CONVERTER OPTIMIZATION” to Anthony Correale Jr. et al., U.S. Pat. No. 7,119,578 (Attorney Docket No. YOR920030373US1) entitled “SINGLE SUPPLY LEVEL CONVERTER” to Anthony Correale Jr. et al., both filed coincident with the parent application and to U.S. Pat. No. 7,091,574 entitled “VOLTAGE ISLAND CIRCUIT PLACEMENT” to Anthony Correale Jr., all assigned to the assignee of the present invention.
BACKGROUND OF THE INVENTION
• 1. Field of the Invention
• The present invention is related to integrated circuit (IC) design circuit design and more particularly, to optimizing standard cell design configurations.
• 2. Background Description
• Semiconductor technology and chip manufacturing advances have resulted in a steady increase of on-chip clock frequencies, the number of transistors on a single chip and the die size itself, coupled with a corresponding decrease in chip supply voltage and chip feature size. Generally, all other factors being constant, the power consumed by a given clocked unit increases linearly with the frequency of switching within it. Thus, not withstanding the decrease of chip supply voltage, chip power consumption has increased as well. Both at the chip and system levels, cooling and packaging costs have escalated as a natural result of this increase in chip power. For low end systems (e.g., handhelds, portable and mobile systems), where battery life is crucial, net power consumption reduction is important but, must be achieved without degrading performance below acceptable levels. Consequently, power consumption has been a major design consideration for designing very large scale integrated circuits (VLSI) such as high performance microprocessors. In particular, increasing power requirements run counter to the low end design goal of longer battery life. Since chip power is directly proportion to the square of supply voltage (V_dd), reducing supply voltage is one of the most effective ways to reduce the power consumption, both active and standby (leakage) power, which is becoming more and more of a problem as technology features scale into nanometer (nm) dimension range.
• While reducing supply voltage is attractive to reduce the power consumption, reducing V_ddincreases transistor and gate delay. Thus, for a design that is performance constrained, the supply voltage may not be lowered too much and, it is usually determined by the most timing critical paths. However, it is often the case that most cells in a chip are timing non-critical. If those timing non-critical cells are properly selected to be on lower supply voltage(s), significant power saving may be achieved without degrading the overall circuit performance.
• One approach to reducing power is to use multiple supply voltages each supplying different circuit blocks or voltage islands. Each voltage island runs at its minimum necessary supply voltage. However, multiple supply voltages on the same circuit/chip present numerous problems, especially for deep submicron (DSM) designs, where circuit performance often is dominated by interconnect delays. In particular, logic synthesis is very complicated for multiple supply designs and, placement and routing must be considered together for voltage assignment, level converter insertion and minimization, and for circuit block clustering to simplify power routing of multiple supply lines.
• Thus, there is a need for circuit element clustering for minimum power and to simplify power routing of multiple supply lines.
SUMMARY OF THE INVENTION
• It is a purpose of the invention to improve integrated circuit (IC) chip design;
• It is another purpose of the invention to improve cell placement in multi supply voltage IC chip designs;
• It is yet another purpose of the invention to improve cell placement of first supply voltage cells with cells of other supply voltages in multi supply voltage IC chip designs;
• It is yet another purpose of the invention to group circuit cells in a multi-supply design close to their respective power supplies;
• It is yet another purpose of the invention to group circuit cells in a multi-supply design to facilitate timing closure;
• It is yet another purpose of the invention to group circuit cells in a multi-supply design for optimum level converter placement;
• It is yet another purpose of the invention to group circuit cells in a multi-supply design for a minimum number of level converters;
• It is yet another purpose of the invention to group circuit cells in a multi-supply design for efficient level converter placement.
• The present invention relates to an integrated circuit (IC) design, method and program product for reducing IC design power consumption. The IC is organized in circuit rows. Circuit rows may include a low voltage island powered by a low voltage (V_ddl) supply and a high voltage island powered by a high voltage (V_ddh) supply. Circuit elements including cells, latches and macros are placed with high or low voltage islands to minimize IC power while maintaining overall performance. Level converters may be placed with high voltage circuit elements.
BRIEF DESCRIPTION OF THE DRAWINGS
• The foregoing and other objects, aspects and advantages will be better understood from the following detailed description of a preferred embodiment of the invention with reference to the drawings, in which:
• FIGS. 1A-C show different state of the art circuit layouts for multi-supply chips;
• FIG. 2 shows an example of a generic voltage island structure formed according to a preferred embodiment of the present invention;
• FIG. 3 shows an example of a flowchart of a method of generic voltage island optimization for low power with rapid timing closure according to a preferred embodiment of the present invention;
• FIGS. 4A-B show an example of the steps in the logic aware voltage assignment;
• FIGS. 5A-B show an isolated V_ddlcell (e.g.,width1 cell) in the middle of a larger V_ddhisland, optimized by changing such isolated cells back to a V_ddhcell;
• FIGS. 6A-F show before and after level converter placement examples, optimized according to a preferred embodiment of the present invention;
• FIGS. 7A-B show an example of a V_ddlfanin cone for an iterative level converter optimization;
• FIG. 8 shows an example of level converter efficiency measurement flow diagram using V_ddlfanin cone size to iteratively locate and delete least efficient level converters;
• FIGS. 9A-B show before and after examples of level converter optimization effected with logic replacement;
• FIG. 10 shows a flow diagram showing an example of the logic replacement;
• FIGS. 11A-B show before and after examples of replacing a buffer and level converter with a single level converter and adjusting placement to meet design objectives;
• FIG. 12 shows a flow diagram for identifying paired level converters and buffers for optimization.
DESCRIPTION OF PREFERRED EMBODIMENTS
• Accordingly, as described hereinbelow, the present invention provides a versatile and generic multi-supply voltage island circuit structure, wherein different supply voltages are assigned at both macro and cell level within the islands. Unless indicated otherwise for simplicity of discussion hereinbelow, logic cell and gate are used interchangeably and each is a sub-circuit of standard cell design. Further, a standard cell design is taken as having the same height, i.e., row height, for most cells. Abutting cells form circuit rows. Also, typical modern application specific integrated circuit (ASIC) and system on a chip (SOC) designs often have many proprietary macros (known in the art as intellectual property (IP) blocks) mixed with standard cells. A voltage island can be a single cell, an IP block or macro or, a continuous region of cells on the same or adjacent rows that have the same power supply voltage (referred to herein as a high voltage supply or V_ddhand a low voltage supply or V_ddl). An output or source drives a net connecting one or more inputs or sinks to the source and a low/high voltage net connects a low/high voltage source to low/high voltage sinks. Also, although described herein in terms of two (2) supplies description, this is for example only and not intended as a limitation. A person skilled in the art would readily understand how to extended the 2 supply description to multiple supply voltages.
• So,FIGS. 1A-C show different state of the art multi-supply chips with examples of well known circuit island placement, e.g., as described in D. E. Lackey et al., “Managing power and performance for system-on-chip designs using voltage islands”, inProc. International Conference on Computer Aided Design, pp. 195-202, November 2002. In the example100 ofFIG. 1A, voltage islands are only allowed at themacro level102,104, with no fine-grained voltage assignment forcells106,108. For deep submicron (DSM) designs, which have tight performance targets, it may not be possible to switch an entire macro between a normal and a lower supply voltage without incurring an overall circuit performance loss. So it would be more flexible if voltage assignment can be done at cell level to exploit positive slacks. The example ofFIG. 1B shows acircuit block110 with cell level voltage assignment, but at the cost of a restricting the layout to alternating or interleaving pairs of high andlow supply rows112,114.FIG. 1C shows another example120, somewhat unconstrained by the requirement of interleaving entire rows. Instead, in this example eachrow122,124,126,128 may have two areas with different voltages (designated H or L), provided each area occupies either the left or right part of the row. Unfortunately, these voltage island patterns or segregated voltage areas unnaturally constrain voltage assignment and/or reduce placement flexibility. Frequently in a typical modern ASIC/SOC design, non-critical regions are interspersed with critical regions in the same circuit row. Typically available such circuit structures are not flexible enough to allow circuit placement or voltage island granularity sufficient to meet stringent delay constraints or, in placing to meet such constraints introduce routing problems.
• By contrast, a preferred circuit and chip design method incorporates generic voltage islands with much finer layout granularity. Supply voltage assignment may be done at both macro and gate level, affording designers much more design freedom and providing a much more flexible voltage island layout structure. Further such a preferred embodiment design achieves timing closure on design timing goals during voltage island formation and hastens timing optimization.
• FIG. 2 shows an example of a generic voltage island structure130 formed according to a preferred embodiment of the present invention, wherein different voltages are assigned at both macro and cell levels. Preferred voltage assignment affords more freedom in terms of layout style by allowing multiple voltage islands within the same circuit row. Further, such a pattern130 is achievable with minimum disturbance to an existing placement, i.e., after normal chip design and placement. So, after designing and placing circuits for performance, for example, the design may be modified according to the present invention, selectively replacing higher power (V_ddh) circuits (stippled) with lower power (V_ddl) circuits (clear) where possible. Since some gap may be needed between adjacent V_ddlislands132 and V_ddhislands134 (depending on the standard cell library), a minimum or maximum allowed cluster size or number of voltage islands may be specified for each circuit row, e.g.,136, based on the particular user or technology specification. See, for example, U.S. application Ser. No. 10/387,728 (Attorney Docket No. RPS9-2002-0253) entitled “VOLTAGE ISLAND CIRCUIT PLACEMENT” to Anthony Correale Jr., filed Mar. 13, 2003, assigned to the assignee of the present invention and incorporated herein by reference. To facilitate power routing, a power grid structure ofVDDL138 andVDDH140 is co-designed with the voltage island assignment.
• Typically, a V_ddlsource cannot drive a V_ddhsink reliably without excessive leakage. Thus, a level converter is needed for a transition from a low voltage net to a high voltage net. Traditional level converters require both supply voltages, V_ddland V_ddh, to avoid excessive leakage. Previously, using dual-supply voltage level converters, required that they be placed at theisland132,134 boundaries for access to both power supplies. However, a single-supply level converter is used such as is described in U.S. Pat. No. 7,119,578 (Attorney Docket No. YOR920030373US1) entitled “SINGLE SUPPLY LEVEL CONVERTER” to Anthony Correale Jr. et al., filed coincident with the parent to this application and incorporated herein by reference. Correale Jr. et al.level converters144 can be placed anywhere in ahigher voltage island134 orlogic146 and so, provide additional placement flexibility. Preferably, a level converters as described hereinbelow is a single supply level converter such as Correale Jr. et al.
• FIG. 3 shows an example of aflowchart150 of a method of generic voltage island optimization for low power with rapid timing closure according to a preferred embodiment of the present invention. For deep submicron (DSM) designs, interconnect delay can dominate the transistor delay, thus placement (and even routing) information are used to get an accurate timing estimation.
• So, beginning instep152 an input netlist description and specifications (e.g., technology files and timing constraints) is provided. In step154 a timing closure tool with Spice RC delays (e.g., a suitable tool from Synopsis, Inc., or EinsTimer from IBM Corporation) is used to determine the entire circuit/chip timing at the higher supply voltage (V_ddh) for a base placement and optimization, i.e., determining global placement and obtaining a good timing estimation. Then, non-critical cells are identified and assigned a lower supply voltage (V_ddl). As noted hereinabove, interconnect delay can dominate the gate delay for deep submicron circuits and so, power can be reduced for lightly loaded circuits where power is not needed for driving large interconnect loads. So, the global placement information is used to correctly identify the critical versus non-critical cells, e.g., heavily loaded verses lightly loaded. Then instep156, a logic aware voltage assignment is performed, assigning the lower supply voltage(s) to less critical circuits, i.e., macro, latch and/or cell. Next, instep158 level converters are inserted and the results are refined and optimized. A level converter is inserted wherever there is a transition net with a low voltage cell driving a high voltage cell or, where a pass gate data input to a low voltage cell or circuit element is being driven by a high voltage cell and being controlled by a low voltage cell. Instep160 isolated assignments are removed in a physical aware voltage reassignment step, locating and reverting solo or very small groups of low voltage circuits that are difficult to form into low voltage islands. Since eliminating those isolated low voltage cells may create opportunities to reassign previously assigned high voltage cells to low voltage cells, instep162 the design is checked for such opportunities. If any are found, returning to step156 for another pass the design is further optimized, until there is no improvement available instep162. Finally, in step164 placement and power routing patterns are effected based on the voltage island assignments to form the final high and low voltage islands. As result, the entire flow can be tightly integrated with a suitable physical synthesis engine166 such as a routing tool from Cadence Design Systems, e.g., for application of any necessary further timing optimization.
• In addition to identifying circuits for separation into voltage islands, supply high and low voltages may similarly be selected to achieve optimum power saving. Further, a preferred voltage assignment method has application to static and incremental timing engines. Every time a macro or cell is changed from a higher voltage cell to a lower voltage cell, or vice verse, the timing (slack) is updated.
• FIGS. 4A-B show an example of the steps in the logic awarevoltage assignment step156 ofFIG. 3. Essentially, a logic assessment is done for each macro1560,latch1562 andcell1564 to determine which may be replaced with a low voltage equivalent and level converter, if required. For checking combinational logic cells instep1564, the cells may be sorted according to timing order from timing end point to timing starting point, i.e., from PO to PI or latch input to latch output. In eachmajor step1560,1562 and1564, each circuit element of each group (macro, latch or cell) is checked, essentially according to the steps1570-1576 inFIG. 4B to identify low voltage candidates. First instep1570, the supply to the macro, latch or cell is reduced and one or more level converters are inserted where appropriate, i.e., at transition nets with low voltage sources driving high voltage sinks. Instep1571 an appropriate incremental timing report is checked for the macro, latch or cell. Then, instep1572, if the timing specification of the macro, latch or cell is met, it is designated to the low supply voltage. For latches in particular, a latch is designated a low supply latch, if all input pins still have positive slack (i.e., edges arrive at inputs prior to a minimum input set up time) and the output pin slack exceeds a minimum threshold, i.e., for a transitional net the output can accommodate the additional delay for an inserted level converter. Otherwise, instep1573 it is reverted to the normal, higher supply. Instep1574, if additional macros, latches or cells have not yet been checked; then in step1575, the next (macro, latch or cell) is selected and returning to step1570, checking continues. Once, each element of the particular group being checked, i.e., instep1560,1562 or1564, checking proceeds to the next group in1562 or1564, respectively, or ends instep1576. After an initial voltage assignment, the voltage assignment may be further refined, including deleting smaller low voltage supply clusters.
• The initial voltage assignment is not physically aware, i.e., no consideration is given to cell placement. As shown in the example ofFIG. 5A, it is possible to assign an isolated V_ddlcell170 (e.g.,width1 cell) in the middle of a larger V_ddhisland,172A-B,174,176,178. Since such an isolated placement may make it difficult to form uniform voltage islands, an optimum placement is facilitated by changing each suchisolated cell170 back to a V_ddhcell170′ as shown inFIG. 5B. It should be noted that initial assignment of these isolated V_ddlcells may have prohibited considering other V_ddlcells as candidates. Thus, a physical aware voltage reassignment is employed to push more cells to V_ddlwhile minimizing the number of level converters and still meeting the physical timing constraints. So, physical adjacency information is used to facilitate the physical aware voltage reassignment and to guide subsequent voltage assignment.
• Physical awarevoltage reassignment step160 inFIG. 3, basically, includes 2 steps. First, a physical adjacency metric (PAM) is computed for the each V_ddlcell. The PAM(k, d) for each particular V_ddlcell is, the total size (i.e., width) of V_ddlcells within the neighboring k rows, including the cell itself, and within diameter range d. Then, all V_ddlcells with a PAM less than certain threshold are reverted to V_ddhcells. Each reversion may present new opportunities for converting some other V_ddhcells that had not been selected in the initial voltage assignment, e.g., due to slack constraints. So, instep162 ofFIG. 3 logic aware voltage assignment is called again with PAM as an additional metric. Only those cells with PAM larger or equal to the selected threshold may be selected as V_ddlcells. Thus, the logicaware assignment step156 and physicalaware reassignment step162 may be iterated until no further improvement is realized.
• In each iteration level converter placement is optimized instep158 to reduce the total number of level converters, gradually deleting the less efficient level converters. Level converters are necessary for transitions between islands, i.e., at least when a V_ddlsource is driving a V_ddhsink. So, for example, branches to those level converters with a small V_ddlfanin may be eliminated (deleting the level converter and returning the prior cell with a V_ddlcell) or another level converter efficiency metric may be used to select level converters for deletion. Further, since level converters and buffers essentially have the same function and so, can be substituted for buffers, optimizing level converters, simultaneously optimizes buffers. In particular, for any V_ddloutput driving multiple V_ddhinputs (i.e., inputs to multiple V_ddhcells), instead of inserting a level converter for each V_ddhinput, a single level converter is shared, provided that timing and electrical constraints are still met.
• FIGS. 6A-F show before and after level converter placement examples. In the example ofFIG. 6A, a V_ddldriver180 is shown driving a transition net with two V_ddhreceivers182,184 aligned in a straight line, where thelevel converter186 is at the geometric center of the tworeceivers182,184. However, this placement increases the total wire length because of the detour from thedriver180 to thelevel converter186 and, then to theleft receiver182. By contrast, as shown inFIG. 6B, an optimized placement places thelevel converter186 just in front of theleft receiver182 to minimize the total net power by maximizing the low voltage net length portion. Thus, power saving may not necessarily decrease the total wire length, but optimizes its apportionment.
• Similarly, as shown in the examples ofFIGS. 6C-D, placement can be optimized for adriver190 driving a transition net withreceivers192,194,196,198 on a two dimensional plane from thedriver190. In this example, thereceivers192,194,196,198 are all located in the first quadrant from the perspective of thedriver190. Acommon level converter200 can be shared between V_ddland V_ddhinterfaces. Preferably, however, theoptimum level converter200 placement is a location to minimize the total wire length; and also, allocates the largest portion of that wire length to the low supply voltage side (i.e., driven by the V_ddldriver190) to minimize switching power, i.e., power expended driving the wire load. Thus, in the example ofFIG. 6C thelevel converter200 is located a minimum power point at (X_min, Y_min), where X_minand Y_minare the minimum x and y coordinates of allreceivers192,194,196,198. Thus selecting the minimum power point avoids any total wire length increase, but may place thelevel converter200 closer to thedriver190. Alternatively, inFIG. 6D thelevel converter200 may be placed at the Manhattan distance from the nearest sink (receiver194 in this example) to the source (on the 45° dottedline202 in this example). A weightedgeometric center204 may be determined for all thereceivers192,194,196,198 from a delay neutral drive point from thelevel converter200. The weight applied for eachreceiver192,194,196,198 is a measure of how close the receiver should be to the driver190 (e.g., the weight may be measured by the slack at each receiver). Then, a projection is determined from the weightedgeometric center204 to the 45° dottedline202 is performed to determine the level converter location. The weighted center placement more aggressively pushes thelevel converter200 further away from thesource190 to increase the total V_ddlwire length and thus reduce V_ddhwire length, and as a result, minimize power.
• FIGS. 6E-F show after placement examples, wherein V_ddhreceivers210,212,214,216 are located in more than just a single quadrant, e.g., they occupy both the first and the fourth quadrant. In this example, thelevel converter218 is placed at a side drive point (X_min, Y_drv), where X_minis the minimum x-coordinate of all receivers, and Y_drvis the y-coordinate of thedriver220. Similar drive points can be located for first-second quadrants, second-third quadrants, and third-fourth quadrants. However, if as in the example ofFIG. 6F, thereceivers230,232,234,236238 are dispersed in diagonal quadrants (e.g., first-third quadrants, or second-fourth quadrants), thelevel converter240 is placed near thedriver242 because it may not be inserted at any other place without increasing the total wire length.
• It should be noted that in all of the above examples, if onelevel converter186,200,218,240 is not enough to drive all the respective V_ddlreceivers, it may be powered up using any suitable technique, e.g., cloning. Whether the level converter is powered up through cloning or otherwise should be evaluated together with the overall power saving of the placement. In particular, the original assignment of V_ddldriver may be reverted to V_ddhif the level converter cost is higher than the gain by selecting the driver to be V_ddlin the first place. Furthermore, level converter placement as described with reference toFIGS. 6A-F is done focusing on total power saving, by minimizing the overall capacitance and V_ddhcell load capacitance, while maximizing the V_ddlcell load capacitance after level converter placement. However, application of the above described level placement may be done guided by any other selected cost function, such as timing and power supply adjacency, i.e., to deliver proper power supplies to level converters. After the level converter is inserted and placed, a Steiner tree is constructed to connect the level converter with the V_ddhreceivers.
• FIGS. 7A-B show an example of an iterative optimization of level converter placement for a V_ddlfanin cone250 according to a preferred embodiment of the present invention. Generally, a fanin cone for level converter includes all gates that drive nets leading to the gate inputs and, as applied to the level converters, signals originating from V_ddlgates without crossing/passing through any V_ddhgates. As a rule of thumb, the larger the V_ddlfanin cone, the more effective the level converter.
• In this example the V_ddlfanin cone250 forlevel converter252 includes the 5gates254,256,258,260,262. In this example, the size of each V_ddlfanin cone for thelevel converters252,266 and268 is 5, 1 and 4, respectively. However, since eachlevel converter252,266,268 consumes power and chip area, placement is optimized by deleting inefficient level converters. To the first order, the size of V_ddlfanin cone is a rough measure of the efficiency of a particular level converter. So, level converters that are inefficient, i.e., level converters with small fanin cones, are deleted. For example, thelevel converter266, which has V_ddlfanin cone size of one (i.e., only onebuffer270 driving into it) and so, may not be cost effective with respect to power or area. Further, as shown inFIG. 7B after deletinglevel converter266 and reverting the single, lowvoltage input buffer270 to V_ddhbuffer272, the inefficient fanin cone has been eliminated. Also, after deletinglevel converter266, the V_ddlfanin cones oflevel converters252 and268 are 4 and 4, respectively.
• FIG. 8 shows an example of level converter efficiency measurement flow diagram280 using V_ddlfanin cone size to iteratively locate and delete least efficient level converters according to a preferred embodiment of the present invention. First, instep282 the V_ddlfanin cone of each level converter is determined. Then, instep284 level converters with a fanin having a cone size less than or equal to a selected threshold, k, are converted to V_ddlcells. Next instep286 the V_ddlfanin cone size for remaining level converters is updated. Instep288 fanin cones are checked to determine whether more inefficiently placed level converters can be removed, i.e., have a fanin cone size below k. If more fanin cones with a size below k remain, then, returning to step284, remaining such inefficient level converters are removed, one at a time until none are found instep288 and optimization ends instep290. Further, a minimum threshold of V_ddlfanin cone size km may be obtained, incrementally, or a total level converter number upper bound may be incrementally increased to gradually reach an optimum placement. So, the bound may be incrementally increased, gradually removing least efficient level converters, i.e., by setting k=1 first, then k=2, 3, and so on until k=k_minor until a selected total level converter number requirement is met. It should be noted also that using V_ddlfanin cone size as described herein as a level converter efficiency metric is for example only and not intended as a limitation. Any other measurement metric may be employed to iteratively select and delete less efficient level converters.
• FIGS. 9A-B show before and after examples,300,302, respectively, of level converter placement optimization effected with logic replacement, i.e., replacing selected V_ddhgates with its V_ddlcounterparts (possibly using a different size in the library) to reduce the number of level converters. In particular, this is effective for those V_ddhgates that have many fanin signals originating with level converters. So for example, in beforecircuit300gate304 is assigned to V_ddh, because it is timing critical due to another input from a V_ddhgate306. Thegate304 receives its four other inputs fromgates308,310,312,314 that are all V_ddlcells and so, require insertion of fourlevel converters316,318,320,322. Thus, in optimizedcircuit302,gate304 is replaced with a functionally equivalent V_ddlgate324 and, typically, a level converter (not shown) is inserted atoutput326. In addition, the replacement V_ddlgate324 may be of a different drive strength. However, the number of level converters may be significantly reduced by such replacement.
• FIG. 10 shows a flow diagram showing an example of thelogic replacement step330 according to a preferred embodiment of the present invention. First, in step332 a V_ddhgate candidate with multiple input level converters is identified. Then, instep334 the selected V_ddhgate is temporarily replaced with its V_ddlequivalent. Unnecessary level converters are deleted from the inputs to the replaced gate and, if necessary, a level converter is inserted at the gate output. Then instep336, the timing constraint is checked to determine if it is still met. Optionally,step334 may be repeated, trying different V_ddlgate sizes and selecting the best result for timing/power. If timing is met instep336, then the logic replacement with the most power saving is selected instep338. Otherwise, instep340 the previous (original) solution is restored. Instep342 the logic is checked to determine if more V_ddhcandidates remain. If so returning to step322 the next V_ddhcandidate is selected, until instep342 no candidates remain and so, all candidate V_ddhgates with multiple level converters in its inputs are checked.
• FIGS. 11A-B show before and after examples350,352, wherein abuffer354 andlevel converter356 are replaced, with asingle level converter358 and placement is adjusted to meet design objectives. As noted hereinabove, since each level converter is itself a buffer, level converters can be substituted for traditional buffers, e.g., as signal relays to break long interconnects and restore/redrive signals, thereby reducing buffers or chains of inverters.
• FIG. 12 shows a flow diagram360 for identifying paired level converters and buffers for optimization. First instep362, a each level converter is identified with at least one buffer immediately before it with fanout1 (FO1). If such a level converter is identified, then instep364 the buffer is temporarily removed, and the level converter placement is adjusted as described hereinabove. Then instep366, the timing specification is checked and, if still met, the buffer is permanently removed. Otherwise, instep368, the original placement is restored. Then, instep370 the remaining buffers are checked for more candidates and, if one is found, returning to step364, that candidate is checked. Otherwise, checking ends instep372.
• A design may be constrained wherein portions may not be modified, e.g., with input/output (I/O) constraints that may not be replaced, for example, with V_ddlcells. For example in a microprocessor core design, placing slower V_ddlcells at the input logic between primary chip input and the first level latches, as well as at the output logic between the final level latches and the primary chip outputs may be unacceptable. Such constrained logic can be hidden or removed from consideration to avoid changing those cells to V_ddlcells. Also, a user may specify a supply voltage for a set or sets of cells or macros. Such constraint information can be passed to voltage assignment with those constrained cells marked as hidden and so, not touched. Also, circuitry related constraints, can be applied during voltage assignment.
• Advantageously, the present invention provides a flexible, systematic method for identifying cell candidates and creating optimized voltage islands. Further, such a design is achieved with a fine-grained voltage island and without performance degradation. Additionally, voltage assignment is both logically and physically, honoring both logic and physical adjacencies. Level converters are efficiently optimized for the design.
• While the invention has been described in terms of preferred embodiments, those skilled in the art will recognize that the invention can be practiced with modification within the spirit and scope of the appended claims.

Claims (1)

1. An integrated circuit (IC) comprising:

a plurality of circuit rows, at least one row of said plurality of circuit rows including three or more voltage islands;

at least one low voltage island in said at least one row, circuit elements in each said at least one low voltage island being powered by a low voltage (V_ddl) supply; and

at least one high voltage island in said at least one row, circuit elements in each said at least one high voltage island being powered by a high voltage (V_ddh) supply, V_ddhbeing a higher voltage than V_ddl.

Images