Patent Application
20090044033
The present invention relates generally to the design of ICs using a gated clock design technique, and more particularly to the low power IC design.
The following documents are incorporated herein by reference, in their entirety, for their useful technical descriptions related to the background explained below: U.S. patent application publication 20030033580 of Maxwell, et al. dated Feb. 13, 2003; U.S. patent application publication 20060220721 of Nitin, et al. dated Oct. 5, 2006; U.S. patent application publication 20070011643 of Qi; et al. dated Jan. 11, 2007; U.S. patent application publication 20070094623 of Haizhou, et al. dated Apr. 26, 2007; U.S. patent application publication 20070099314 of Haizhou, et al. dated May 3, 2007; U.S. Pat. No. 6,810,482 to Saxena, et al. dated Oct. 26, 2004; U.S. Pat. No. 7,007,247 to Wang, et al dated Feb. 28, 2006; U.S. Pat. No. 7,051,306 to Hoberman, et al. dated May 23, 2006; and U.S. Pat. No. 7,076,748 to Kapoor, et al. dated Jul. 11, 2006.
In recent years, the size of integrated circuits (ICs) has dramatically increased in both size and number of transistors, resulting in higher power consumption. In typical IC designs, the clock distribution network, i.e., the clock tree, can consume from twenty to fifty percent of an IC's total active power. One important technique for reducing power consumption in IC designs is to reduce the power of an IC's clock distribution tree by gating portions of the IC that do not need to be clocked under certain conditions.
This process, known as “clock gating”, disables the clocks fed to logic blocks of the IC when the logic blocks are not currently enabled or otherwise in active use. Power consumption due to the clocking of logic blocks that are not directly involved with the current operation of the IC is thereby minimized.
Furthermore, gating all registers in the design increases the size of the IC. As a result, traditional approaches gate all the registers having a size that is greater than a predefined threshold. Alternatively, registers to be clock-gated are manually selected. These approaches are not optimal in terms of power and area cost, as most of the savings can be achieved using only a few clock gating structures.
One solution for selection of the optimal registers to be clock-gated is based on computing the activity savings per each such register. This solution is described in a U.S. patent application Ser. No. 11/419,624 by Kapoor, et al. assigned to common assignee and which is hereby incorporated by reference in its entirety. However, this solution does not compute the power savings for the different clock gating techniques, such as those described above. Furthermore, the computation is limited to clock gates (or “enables”) already existing in the design.
It would be therefore advantageous to provide a more efficient approach for computing the power savings of candidate registers for clock gating implementation.
The present invention includes a method for computing the power savings in clock gating circuits in integrated circuit (IC) design. The method computes the difference in power savings between techniques used for clock gating. Based on the computation results, the method outputs a script to control an implementation tool causing the use of the best implementation clock-gating technique in terms of power and area savings. Trying to compute the power savings manually by a designer is not feasible, as a typical design may include hundreds of candidate clock gating registers.
It will be appreciated that, in the present description, the concept of using “the best” implementation does not mean that the implementation is optimal for every possible criterion. The approach described below allows for the automated selection of an implementation among alternative implementations. It is assumed that the needs of the specific situation facing a person in this field will dictate what is “best”. With this in mind, the concept might be restated as that of selecting an implementation on the basis of a particular criterion or of particular criteria. Another way to put this is to say that a preferred alternative is selected. In the description, the criterion is that of the greatest power savings value.
At S305, a code representing a RTL description of an IC design is received. The code may be written in any hardware description language (HDL) including, but not limited to, Verilog, VHDL and the like. At S310, a synthesized netlist is produced by an IC synthesis tool. Synthesis tools produce a gate level netlist based on a RTL representation. The netlist generally include logical gates such as AND, NAND, NOR, OR, XOR, NXOR, and the likes. One such synthesis tool is described in the US patent entitled “An Apparatus and Method for Handling of Multi-Level Circuit Design Data”, patent number 6,993,733, assigned to common assignee and which is hereby incorporated by reference in its entirety. At S320, a list of candidate registers (hereinafter the “input candidate list”) in the design to be clock-gated is received. At S330 a process for detecting new candidate registers in addition to those provided at S320 is executed. Specifically, the process detects registers that can be clock-gated without changing the design functionality, and outputs a complete list of candidate registers to be clock gated (hereinafter the “output candidate list”).
The present invention is concretely described in terms of several embodiments for detecting new candidate registers (or enables). In one embodiment the process traces forward in combinational logic to detect candidates. As shown in
In accordance with another embodiment of the present invention the process traces backward in combinational logic. As shown in
In another embodiment the process traces backwards in a sequential logic to detect new candidates. As illustrated in
At S340, for each register in either in the input or the output candidate list, the power savings is computed. Refer now to
At S510, for each clock signal in the design a virtual buffer may be created. The virtual buffer is a virtual clock tree, which is an estimation of the power and area required by a real clock tree. The clock tree is inserted at a later stage in the design process by a clock tree synthesis tool. The fanout is the output terminals on a clock signal or the net attached to an output terminal. In case the number of registers on the clock net is low, then a normal buffer can drive the load and no virtual buffer is needed; the virtual buffer is required when the fanout is too high for a normal buffer to drive the load. The fanout limit beyond which a normal register can drive is given in the technology library. The technology library includes details on fabrication processes. The same design implemented in two different fabrication processes may have different power consumption and different clock gating tradeoffs. At S520, a single register is selected and removed from the candidate list. At S530 the gating technique that the designer utilized to gate the enable signal of the selected candidate register is determined. That is, the method checks, if the selected register has been gated by the designer using an ICGC or a multiplexer (MUX). As the objective of the method is to compute the difference in power between the two techniques, the method prepares to calculate a power savings for the selected register by inserting virtual logic. Specifically, the method continues with S540 to inset a virtual ICGC; otherwise, at S545 a virtual MUX is added to the circuit. It should be noted that the additional logic (MUXs, ICGCs, or virtual buffers) is added merely for the power computing and removed thereafter.
As an example,
At S550, the power savings (PS) for the selected candidate register is calculated. In accordance with one embodiment of the present invention the PS is determined according to the following equation:
PS=(DynGCV*ActClk/ActEnab)+PwrVmux−PwrAnd;
where,
ActClk is the activity of clock signal (clk);
ActEnab is the activity of enabled clock;
DynGCV is the dynamic power of enabled clock virtual buffer (e.g., buffer 640);
PwrAnd is the total power of the ICGC; and
PwrVmux is total power of all virtual MUXes.
The activity of the clock and enable signal (ActClk and ActEnab) values are measured by reading input simulation data. The activity statistic is defined as the average number of times that a signal changes value during a simulation period, divided by the number of clock cycles in the simulation period. The DynGCV, PwrVmux, and PwrAnd values are the result of a power calculation process, which typically computes the power using a netlist, simulation data, and a power library. The power library is part of the technology library. The power calculation for the virtual MUX uses activity data of the input signals (d), output signals (q) and an enable signal of the candidate register.
At S560 a check is performed to determine if the input and output candidate lists are empty, i.e., if the power savings is computed for all registers in that list, and if so execution ends; otherwise, execution returns to S520 were another register from the list is selected.
Referring back to
It should be noted to a person skilled in the art that methods and processes described herein can be implemented in software, hardware, firmware, or combination thereof. The implementation may be performed as well using a computer system having a processor and a memory under control of the processor, the memory storing instructions adapted to enable the processor to carry out operations as described above. The implementation may be realized, in a concrete manner, as a computer program product that includes a tangible computer readable medium holding instructions adapted to enable a computer system to perform the operations as described above. The computer program product may optionally be a CAD program, and the computer system may optionally be a CAD system.
The foregoing exemplary embodiments, described in general terms above, are commended to the person familiar with this field. Those not familiar with this field may make recourse to the several helpful background references mentioned earlier in this description. The person familiar with this field will appreciate that the invention is more general than the concrete embodiments described above, and will find the scope of the invention to be described by the appended claims. In addition, the person familiar with this field will understand that various modifications may be made without departing from the scope and spirit of the invention, such as inserting steps in addition to those mentioned, performing the steps in various different orders or contemporaneously where possible, combining or omitting various steps, or the like.
