Logic synthesis method, semiconductor integrated circuit and arithmetic circuit

Abstract

A top-down design technique is used to design a semiconductor integrated circuit having a plurality of registers and a plurality of combinational circuits each of which is connected between the registers. When a semiconductor integrated circuit is logic-synthesized from a register transistor level, a front section of a critical path-containing combinational circuit is driven by a high voltage from a high-voltage source while the remaining section and other combinational circuits with no critical path are driven by a low voltage from a low-voltage source. A level shifter is placed at the stage before the critical path-containing combinational circuit. The level shifter converts a low-voltage signal into a high-voltage one. This invention facilitates logic synthesis of a low-power semiconductor circuit without increasing the maximum signal propagation delay of the critical path and without having to provide a level shifter in a combinational circuit.

Description

BACKGROUND OF THE INVENTION

1. Field of the Invention

This invention relates to an improved logic synthesis method for generating from register transfer levels (RTLs) a semiconductor integrated circuit. Particularly, this invention pertains to a logic synthesis method capable of generating a low-power semiconductor integrated circuit (SIC) and to a low-power SIC made by means of the logic synthesis method.

2. Description of Related Art

Various top-down design techniques have been employed for generating SICs. A top-down design technique represents a target SIC with RTL functional descriptions. More specifically, a target SIC is generated from RTL functional descriptions by means of logic synthesis.

FIG. 29 shows RTL functional descriptions. FIG. 30 shows a logic circuit, i.e., an SIC, generated from the RTL functional descriptions.

The functional descriptions specify respective register-to-register data transfers, at functional level. r1, r2, r3 and r4 are registers. func1, func2, func3 and func4 are functional descriptions of combinational circuits (CCs) each of which is connected between the registers. assign is a sentence descriptive of a connection relationship between a signal and a CC. Also, always is a sentence descriptive of a connection relationship between a register and a signal.

When synthesizing a logic circuit from the RTL functional descriptions shown in FIG. 29, the circuit is determined on the area/rate tradeoff curve by giving a constraint of the area and a constraint of the rate.

FIG. 30 shows the resulting logic circuit. 101, 103, 105 and 107 are flip-flop (FF) circuits mapped from r1, r2, r3 and r4 by means of logic synthesis. There is a direct correspondence between the FF circuits (101, 103, 105, 107) and the registers (r1, r2, r3, r4). 108 is a clock buffer. 100, 102, 104 and 106 are CCs and there exists a correspondence between these CCs and func1-func4. The CCs 100, 102, 104 and 106 are circuits that are mapped, as circuits on the area/rate tradeoff curve, from the FIG. 29 RTL functional descriptions.

The power consumption of an SIC, P, can be found by the following equation.

P=f.times.C.times.V.sup.2 where f is the operating frequency, C is the load S capacitance and V is the supply voltage. There are three possible ways of reducing P. The first is to reduce f. The second is to reduce C. The last is to reduce V. Of these three ways the third one is considered the most attractive way since it is able to the most effectively reduce the power consumption.

However, there is the problem that as the supply voltage V decreases, a critical path of a great number of paths together forming a logic circuit, which has the maximum delay of all the paths, comes to have a greater delay.

With a view to providing a solution to the foregoing problem, Japanese Patent Application, published under Pub. No. 5-299624, proposes a technique. In accordance with this technique, low-speed logic gates, not required to operate at a high speed, are driven by a low-voltage source, whereas high-speed logic gates, required to operate at a high speed, are driven by a high-voltage source. This technique is trying to achieve a low-power SIC by driving logic gates forming a critical path by a high-voltage source and by driving other logic gates by a low-voltage source, with no increase in critical path's delay. This approach, however, produces the following drawbacks.

Suppose data is transferred from a slow logic gate driven by a low-voltage source to a fast logic gate driven by a high-voltage source. In this case, there is a need to arrange between these two logic gates a level shifter capable of converting a low-level output from the slow logic gate to a high level (see Japanese Patent Application, published under Pub. No. 5-67963). However each of the CCs shown in FIG. 30 is made up of great many logic gates (see FIGS. 31 and 32). In order to drive a critical path, represented by bold line, by high voltage, level shifters must be arranged at points marked with circles. In FIG. 31, a required number of level shifters is eight. In FIG. 32, twelve level shifters must be placed. A high-density SIC includes a great number of CCs, therefore containing a great number of logic gates. Accordingly, a critical path-containing CC of the high-density SIC requires a great number of level shifters. In addition, there are many CCs with a critical path. This means that the placement of an enormous number of level shifters is required. In high-density SIC design, it may be possible to determine where to place level shifters in a particular CC. However, it becomes considerably troublesome to determine where to place level shifters in an entire SIC. The placement of level shifters is therefore time consuming and design work becomes most complicated.

SUMMARY OF THE INVENTION

Bearing in mind the above-mentioned problems with the prior art techniques, the present invention was made. It is therefore an object of the present invention to provide an improved logic synthesis method capable of facilitating low-power SIC (semiconductor integrated circuit) generation, with no increase in the delay time of critical paths contained in CCs (combinational circuits) of an SIC. It is another object of the present invention to provide a low-power SIC, with no increase in the critical path delay time.

By directing attention to the following points, this invention was made. An SIC, shown in FIG. 30, is formed by many registers and many CCs connected between the registers. If it is arranged so that a register contains a level shifter, this eliminates the need for placing a level shifter at plural locations of CCs at which the provision of level shifters is required at the time of driving a critical path by high voltage. This cuts down the number of level shifters required.

Statistically, it is said that in an SIC, the number of logic gates arranged along critical paths is about 5% of the total number of logic gates contained in the entire SIC. Therefore, the ratio of the number of CCs having a critical path versus the total number of CCs contained in the entire SIC is negligible. Even if, as illustrated above, a register is provided with a level shifter and the entirety of a critical path-containing CC that receives data from the level shifter is driven by high voltage, this will not result in a great increase in power consumption.

It is sufficient that the critical path maximum delay is held below a design delay upper limit. The entire CC with a critical path is not necessarily driven by high voltage. By driving a part of the CC by high voltage, it becomes possible to reduce the maximum delay time to below the design delay upper limit and, as a result, the power consumption can be held as low as possible.

As a general rule, a level shifter is arranged only in a register and only a part of a CC having a critical path is driven by a high-voltage source.

The present invention provides a logic synthesis method for synthesizing, based on interconnection information of logic cells, a semiconductor integrated circuit having a plurality of registers and a plurality of combinational circuits each of which is connected between the registers. This logic synthesis method comprises:

(a) a first step that includes:

mapping, when there exists a combinational circuit in said plural combinational circuits that has a signal propagation delay not more than a design delay upper limit, said combinational circuit into a combinational circuit of a first type driven by a low voltage from a low-voltage source; mapping, when there exists a combinational logic circuit in said plural combinational circuits that has a signal propagation delay above the design delay upper limit, a part of said combinational circuit into a combinational circuit of a second type driven by a high-voltage from a high-voltage source, and mapping the remaining part of said combinational circuit into a combinational circuit of the first type in order that said combinational circuit has a signal propagation delay below the design delay upper limit;

(b) a second step that includes:

checking if said first step produces a mixture of a combinational circuit of the first type and a combinational circuit of the second type arranged such that an output from said first-type combinational circuit serves as an input to said second-type combinational circuit; if such a mixture is found to exist, remapping said first-type combinational circuit into a combinational circuit of the second type;

(c) a third step that includes:

determining which of said plural registers provides a signal to said mapped second-type combinational circuit or to said remapped second-type combinational circuit; mapping a register of said plural registers that is found to provide a signal to said second-type combinational circuit into a register driven by a voltage from a voltage source including said high-voltage source, and mapping a register of said plural registers that is found to provide no signal to said second-type combinational circuit into a register driven by a low voltage from said low-voltage source.

The present invention is characterized in that the first step includes:

conceptually dividing, if there exists a combinational circuit of the first type that has a signal propagation delay above the design delay upper limit, said first-type combinational circuit into plural combinational circuit elements, and remapping the first of said plural combinational circuit elements into a combinational circuit of the second type; checking if said post-remapping combinational circuit has a signal propagation delay above the design delay upper limit; remapping, when said post-remapping combinational circuit is found to still have a signal propagation delay above the design delay upper limit, the second combinational circuit element, located next to the aforesaid first combinational circuit element in the direction of signal propagation in said first-type combinational circuit, into a combinational circuit of the second type, checking if said post-remapping combinational circuit of the second type has a signal propagation delay above the design delay upper limit, and repeating said remapping/checking processing.

The present invention is characterized in that:

said first step including:

(i) setting in a combinational circuit of combinational circuit elements that has a signal propagation delay above the design delay upper limit a plurality of windows having a specified size, each said window serving as a search range; (ii) selecting from among said plural windows a window that contains a group of combinational circuit elements of said elements whose total area or total delay is the smallest, and mapping said group as a part of said combinational circuit into a combinational circuit of the second type; (iii) placing a level shifter at a stage before said mapped second-type combinational circuit.

The present invention provides a semiconductor integrated circuit comprising a plurality of registers and a plurality of combinational circuits each of which is connected between said registers wherein:

a part of said plural combinational circuits are formed of combinational circuits of a first type driven by a low voltage from a low-voltage source; an internal section of each of the remaining part of said plural combinational circuits is formed of a combinational circuit of a second type driven by a high voltage from a high-voltage source and the remaining internal section thereof is formed of a combinational circuit of the first type; of said plural registers one with a combinational circuit of the second type on its output side is formed of a register driven by a voltage from a voltage source including said high-voltage source.

The present invention provides an arithmetic circuit comprising a plurality of arithmetic element rows each of which contains a given number of arithmetic elements wherein (A) arithmetic elements contained in the first of said plural arithmetic element rows each receive an external signal, (B) arithmetic elements contained in the remaining other arithmetic element rows each receive an output from arithmetic elements contained in its preceding arithmetic element row, and (C) arithmetic elements contained in the last of said plural arithmetic element rows each provide an arithmetic operation result,

said arithmetic elements contained in the last arithmetic element row being driven by a high voltage from a high-voltage source; said arithmetic elements contained in the remaining other arithmetic element rows except the last arithmetic element row being driven by a low voltage from a low-voltage source; a level shifter driven by a high-voltage from said high-voltage source being placed between an arithmetic element of the last arithmetic element row and another of its preceding arithmetic element row so that a low-voltage output signal from said arithmetic element of said preceding arithmetic element row is converted into an output signal having a high voltage from said high-voltage source.

Advantages of the present invention are as follows. A semiconductor integrated circuit of this invention is formed by a plurality of registers and a plurality of CCs connected between the registers, wherein some of the CCs have a critical path. A register located at a stage before a critical-path containing CC, i.e., a register that provides data to the critical path containing CC, is provided with a level shifter. Only a part of the critical path-containing CC is driven by high voltage. The remaining part thereof is driven by low voltage. The other CCs having no critical path are driven by low voltage.

Since a part of a critical path-containing CC is driven by high voltage, this allows the critical path to have a delay below the design delay upper limit. Additionally, a register ahead of a critical path-containing CC is provided with a single level shifter, which makes it possible to further reduce the number of level shifters required in comparison with cases where only critical paths are driven by high voltage. This facilitates the SIC design process. Since the ratio of the number of critical path-containing CCs to the total number of Ccs is negligible and only a part of such a "rare" critical path-containing CC is driven by high-voltage, this holds the increase in power consumption as low as possible. Meanwhile, the remaining part of that critical path-containing CC and the many other CCs without a critical path are driven by low voltage, which greatly reduces the power consumption. As a result, the SIC overall power consumption is reduced.

Additionally, combinational circuit elements (CCEs) forming a CC are mapped one after another into a CCE of the second type in the order in which they receive a signal, in other words, such mapping is carried out in the direction of input signal propagation. As a result of such arrangement, it becomes possible to define an exact boundary between a part of a specific CC and the remaining part thereof (i.e., a boundary between an area of a CC to be mapped into a CC of the first type and an area thereof to be mapped into a CC of the second type). This minimizes the number of CCEs of the second type driven by high voltage, therefore further reducing power consumption.

In accordance with the present invention, a part of a CC to be a CC of the second type is searched using a window. A section, which is found to have either a minimum CCE area (i.e., a smallest number of CCEs) or a lowest processing rate, is designated as a part to be mapped into the second-type CC. This arrangement achieves further reduction of the power consumption and improves the rate of processing.

The present invention provides an arithmetic circuit with many inputs and a few outputs. Only an arithmetic element at the end is driven by high voltage, which, in comparison with a case where a plurality of arithmetic elements on the input side are driven by high voltage, achieves further reduction of the power consumption. Additionally, a level shifter is arranged only in front of the arithmetic element at the end. This, in comparison with a case where a level shifter is arranged in front of each arithmetic element on the input side, requires a less number of level shifters.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 schematically shows the entire structure of an image processor.

FIG. 2 schematically shows the entire structure of a semiconductor chip.

FIG. 3 shows connections between registers and CCs (combinational circuits) of an SIC (semiconductor integrated circuit) made in accordance with an SIC logic synthesis method of a first embodiment of the present invention.

FIG. 4 shows the structure of an FF (flip-flop) circuit having no level shifters.

FIG. 5 shows the structure of an FF circuit having a level shifter.

FIGS. 6a and 6b each show a level shifter structure.

FIG. 7 shows the structure of a latch circuit having no level shifters.

FIG. 8 shows the structure of a latch circuit having a level shifter.

FIG. 9 schematically shows the entire structure of a logic synthesis system.

FIG. 10 shows a set of descriptions in hardware language.

FIG. 11 shows a net list.

FIG. 12 shows a schematic.

FIG. 13 is a flow chart diagram useful in understanding an SIC logic synthesis method in accordance with the first embodiment.

FIG. 14 shows a cell library table.

FIG. 15 shows a part of an SIC.

FIG. 16 shows an SIC with a modified part.

FIG. 17 shows a set of RTL descriptions.

FIG. 18 shows a set of RTL descriptions modified when logic synthesis cannot be executed.

FIG. 19 is comprised of FIGS. 19(a)-(e) wherein FIG. 19(a) describes a first sequence of the first embodiment logic synthesis, FIG. 19(b) describes a second sequence, FIG. 19(c) describes a third sequence, FIG. 19(d) describes a final sequence, and FIG. 19(e) shows an example to be compared.

FIG. 20 is a flow chart diagram useful in understanding an SIC logic synthesis method in accordance with a second embodiment of the present invention.

FIG. 21 is a flow chart diagram useful in understanding an SIC logic synthesis method in accordance with a third embodiment of the present invention.

FIG. 22 is a flow chart diagram useful in understanding an SIC logic synthesis method in accordance with a fourth embodiment of the present invention.

FIG. 23 shows functional descriptions serving as input of a logic synthesis method according to the present invention.

FIG. 24 shows functional descriptions serving as input of a conventional logic synthesis method.

FIG. 25(a) shows the structure of a basic adder, FIG. 25(b) shows an adder generated in accordance with the present invention, and FIG. 25(c) shows an adder generated in accordance with a logic synthesis method to be compared.

FIGS. 26 and 27 shows a flow chart diagram useful in understanding an SIC logic synthesis method in accordance with a fifth embodiment of the present invention.

FIG. 28(a) shows CCEs (combinational circuit elements) within a first window of the fifth embodiment logic synthesis method, FIG. 28(b) shows CCEs within a second window thereof, and FIG. 28(c) shows CCEs within a third window thereof.

FIG. 29 shows a set of RTL descriptions.

FIG. 30 shows a logic circuit of a conventional SIC.

FIG. 31 shows the placement location of level shifters when only critical paths are driven by high voltage in an arbitrary SIC.

FIG. 32 shows the placement location of level shifters when only critical paths are driven by high voltage in another arbitrary SIC.

FIG. 33 shows functional descriptions serving as input of a logic synthesis method in accordance with a sixth embodiment of the present invention.

FIG. 34 is a circuit diagram of a multiplier of the carry save type generated by the sixth embodiment logic synthesis method.

FIG. 35(a) is a graph showing the relationship between signal propagation delay and the number (distribution) of CCs having the signal propagation delay in a conventional technique and FIG. 35(b) is a graph showing the relationship between signal propagation delay and the number of CCs having the signal propagation delay in the present invention.

FIG. 36 shows an SIC generated in accordance with a logic synthesis method of a seventh embodiment of the present invention.

FIG. 37 is a flow chart diagram useful in understanding an SIC logic synthesis method of the seventh embodiment.

FIG. 38 is a flow chart diagram useful in understanding another SIC logic synthesis method of the seventh embodiment.

FIG. 39 shows a result of the seventh embodiment.

FIG. 40 is a flow chart diagram useful in understanding still another SIC logic synthesis method of the seventh embodiment.

FIG. 41(a) is a graph showing the relationship between signal propagation delay and the number (distribution) of CC having the signal propagation delay in a conventional technique and FIG. 41(b) is a graph showing the relationship between signal propagation delay and the number of CCs having the signal propagation delay in the present invention.

FIG. 42 depicts the structure of an inverter that is placed in a register.

DETAILED DESCRIPTION OF THE INVENTION

Referring now to the accompanying drawings, preferred embodiments of the present invention will be described below.

EMBODIMENT 1

FIG. 1 schematically shows an image processor A employing an SIC (semiconductor integrated circuit) based on the present invention. 10 is an analog-to-digital converter (ADC) for converting an external analog signal to a digital output. 11 is a general dynamic random access memory (DRAM). 12 is a first SIC of the present invention capable of reading data from the DRAM 11 or writing data to the DRAM 11 while performing image processing. 13 is a general microcomputer for controlling the first SIC 12. 14 is a second SIC which receives signals from the first SIC 12 to perform image processing.

15 is a source of high voltage (for example, 3V) which is arranged outside the image processor A. 16 is a source of low voltage (for example, 2V) which is arranged outside the image processor A. The image processor A has a high-voltage line 17 coupled to the high-voltage source 15 and a low-voltage line 18 coupled to the low-voltage source 16. In order to achieve reduction of the power consumption of the image processor A, the low-voltage source 16, which is a voltage source for the first and second SICs 12 and 14, supplies low-level voltages through the low-voltage line 18 to the first and second SICs 12 and 14 only. On the other hand, high-level voltages are applied through the high-voltage line 17 to the ADC 10, to the DRAM 11, and to the microcomputer 13. Since a high-level interface voltage is required between circuits of the circuits 10-14, a high-level voltage is also applied to the first and second SICs 12 and 14.

It is possible to replace the low-voltage source 16 with an internal low-voltage source the supply voltage of which results from lowering the voltage of the high-voltage line 17 by means of an internal transistor by its threshold voltage. Such an internal low-voltage source is shown in Japanese Patent Application, published under Pub. No. 4-96369. The externally-arranged, low-voltage source 16 is no longer required.

FIG. 2 depicts the internal structure of the first SIC 12 for image processing. 20 is a chip. 21, 21 are input/output pads provided on the peripheral areas of the chip 20. 22 is an internal core section that is an area other than the areas where the input/output pads 21 are arranged. Provided within the internal core section 22 are five functional blocks A, B, C, D, and E. The functional blocks A-E are arithmetic processing circuits for performing different arithmetic operations. For example, the functional block E is formed of a memory cell section with a small storage capability, e.g., ROM and RAM.

The present invention is applied to the functional blocks A-D of the first SIC 12.

FIG. 3 shows a logic circuit diagram of one of the functional blocks of the first SIC 12 (e.g., the functional block A).

The functional block shown in FIG. 3 is a logic circuit generated from the FIG. 29 RTL functional descriptions by means of logic synthesis. 2, 4, 6, and 8 are FF (flip-flop) circuits respectively forming registers r1, r2, r3, and r4 of the FIG. 29 RTL functional descriptions. Each register r1-r4 is a register forming a part of a pipeline processor of the computer. 1, 3, 5, and 7 are CCs respectively forming func1, func2, func3, and func4 of the FIG. 29 RTL functional descriptions and which are arranged between the registers or at their respective preceding stages. For convenience, the output from each CC is shown in FIG. 3 such that it is applied only to a FF circuit at the next stage. However, the output of a CC may be transferred to another CC.

The FF circuits 2, 6, and 8 are 2V FF circuits driven by the low-voltage source 16. The FF circuit 4 is a 2V/3V FF circuit driven by both the low-voltage source 16 and the high-voltage source 15. Whereas the 2V/3V FF circuit 4 has a level shifter which will be described later, the 2V FF circuits 2, 6, and 8 have no level shifters. Each of the CCs 1, 3, and 7 is formed of a 2V CC (2VCC) of the first type driven by the low-voltage source 16. The CC 5, since it is required to operate at a high speed, is formed such that its front section is comprised of a 3VCC of the second type driven by the high-voltage source 15 and the rear section is comprised of a 2VCC of the first-type driven by the low-voltage source 16.

9 is a 2V clock buffer driven by the low-voltage source 16. This clock buffer 9 is a clock supply means for providing a clock to each FF circuit 2, 4, 6, and 8.

FIG. 4 shows the structure of the FF circuits 2, 6, and 8 having no 2V level shifters. 30 is a master latch that receives an external signal D. 31 is a slave latch coupled in series to the output of the master latch 30. This slave latch 31 provides two signals that are in complementary relationship. The master latch 30 and the slave latch 31, which are connected in series, together form a temporary memory 36. The master latch 30 has an inverter 34a. The slave latch 31 has an inverter 34b. 32 is an output buffer that is coupled to the output of the slave latch 31. 33 is an internal clock generator that generates complementary internal clocks CK and NCK from an external clock signal CLK. These circuits 30-33 are 2V circuits driven by the low-voltage source 16.

FIG. 5 shows the structure of the FF circuit 4 with the foregoing 2V/3V level shifter. The FF circuit 4 has the master latch 30 and the slave latch 31 which are connected in series as in the 2V FF circuit 2, and the internal clock generator 33. The FF circuit 4 further has an output buffer 34 driven by the high-voltage source 15 and a level shifter 35 connected between the slave latch 31 and the output buffer 34. This level shifter 35 is of the 2V/3V type. The level shifter 35 receives from the slave latch 31 complementary signals with a potential difference of 2V (low) therebetween and performs level conversion so that the complementary signals are outputted having a greater potential difference of 3V (high) therebetween.

The structure of the level shifter 35 is illustrated by making reference to FIGS. 6(a) and 6(b). 40 and 41 are PMOS transistors. 42 and 43 are NMOS transistors. The PMOS transistor 40 is connected in series with the NMOS transistor 42 to form an in-series circuit. The PMOS transistor 41 is connected in series with the NMOS transistor 43 to form another in-series circuit. Both the in-series circuits are arranged between the high-voltage source 15 and ground. The gate of the PMOS transistor 40 is connected to the drain of the NMOS transistor 43. The gate of the PMOS transistor 41 is connected to the drain of the NMOS transistor 42. Complementary outputs are extracted from the drains of the NMOS transistors 42 and 43.

As a result of such arrangement, the PMOS transistor 40 and the NMOS transistor 42 jointly act as an inverter, and the PMOS transistor 41 and the NMOS transistor 43 jointly act as an inverter. The slave latch 31 gives complementary outputs so that the gate of the NMOS transistor 43 receives a low voltage of 2V and the gate of the NMOS transistor 42 receives a voltage of 0V. As a result, the NMOS transistor 43 turns on and the NMOS transistor 42 turns off. With such changes in the state, the PMOS transistor 40 turns on and the PMOS transistor 41 turns off. This establishes a connection between the drain of the NMOS transistor 42 and the high-voltage source 15, and the drain of the NMOS transistor 43 is connected to ground, whereupon complementary outputs are produced, with a high potential difference of 3V existing therebetween.

One advantage of the FIG. 6(a) structure is that it is possible to perform level conversion on complementary outputs of the slave latch 31 with a potential difference of 2V therebetween, to produce complementary outputs with a greater potential difference of 3V therebetween wherein neither a through current from the high-voltage source 15 to the low-voltage source 16 nor a through current from the high-voltage source 15 to ground (i.e., 0V) flows.

FIG. 6(b) shows a different level shifter 35'. This level shifter 35' does not have the NMOS transistors 42 and 43 shown in FIG. 6a but employs CMOS inverters 45 and 46 instead. The CMOS inverter 45 is formed by a PMOS transistor 47 and an NMOS transistor 48. The PMOS transistor 47 and the NMOS transistor 48 are connected in series. The CMOS inverter 46 is formed by a PMOS transistor 49 and an NMOS transistor 50. The PMOS transistor 49 and the NMOS transistor 50 are connected in series. The input terminals of the CMOS inverters 45 and 46 (i.e., the gates of the PMOS transistor 47 and the NMOS transistor 48 and the gates of the PMOS transistor 49 and the NMOS transistor 50) receive from the slave latch 31 two outputs in complementary relationship, respectively. The output terminal of the CMOS inverter 45 (i.e., a connection point where the PMOS transistor 47 and the NMOS transistor 48 are connected together) is connected to the gate of the PMOS transistor 41 which is not connected in series with the CMOS inverter 45. The output terminal of the CMOS inverter 46 is connected to the gate of the PMOS transistor 40 which is not connected in series with the CMOS inverter 46. The outputs of the CMOS inverters 45 and 46 is the complementary outputs of the level shifter 35'.

As a result of such arrangement, complementary outputs from the slave latch 31 with a 2V difference therebetween can be level-converted into another pair of complementary outputs with a 3V deference therebetween, wherein neither a through current from the high-voltage source 15 to the low-voltage source 16 nor a through current from the high-voltage source 15 to ground (i.e., 0V) flows. Additionally, the PMOS transistors forming the CMOS inverters 45 and 46 control the flow of through current from the high-voltage source 15 to ground in the transient state.

As described above in FIG. 3, the FF circuit 2, which has the 2VCC 1 on its input side and the 2VCC 3 on the output side, is of the 2V type. The FF circuit 4, which has the 2VCC 3 on its input side and the 3V/2VCC 5 on the output side, is of the 2V/3V type. The FF circuit 6 has the 3V/2VCC 5 on its input side and the 2VCC 7 on the output side, is of the 2V type.

Although in the description made above each register r1-r4 is formed by an FF circuit, the register may be formed by a latch.

Such a latch is illustrated in FIGS. 7 and 8. FIG. 7 shows a low-voltage (2V) latch 51. The latch 51 has a latch section 52 (i.e., a temporary storage), an output buffer 53, and an internal clock generator 54. The latch section 52 inputs and latches a signal D to generate complementary outputs. The output buffer 53 is connected to the output of the latch section 52. The internal clock generators 54 generates an internal clock NG from an external clock G and provides the internal clock NG to the latch section 52. The external clock G is also applied to the latch section 52. These circuits 52-54 are of the 2V type driven by the low-voltage source 16.

FIG. 8 shows a latch 51' of the 2V/3V type. The 2V/3V latch 51' has, in addition to the 2V latch section 52 driven by the low-voltage source 16 and the internal clock generator 54, a 3V output buffer 55 driven by the high-voltage source 15 and a level shifter 56 connected between the latch section 52 and the output buffer 55. The level shifter 56 receives a pair of complementary signals with a low potential difference of 2V existing therebetween to perform a level conversion in such a way as to provide another pair of complementary signals with a high potential difference of 3V existing therebetween. This level shifter 56 is identical in configuration with the one shown in FIG. 6(a) or in FIG. 6(b).

Referring now to FIGS. 9 and 13, a logic synthesis algorithm for the generation of an SIC of FIG. 3 according to logic-cell connection information, is described below.

FIG. 9 shows the entirety of a logic synthesis system 60. 61 is an input (read-in) section. 62 is a translation section. 63 is an optimization section. 64 is a cell mapping section. 65 is a timing verifier. 66 is a circuit diagram generator. 67 is an output section.

The input section 61 puts in either an RTL functional description in HDL (see FIGS. 10 and 29), a net list (see FIG. 11) which specifies, on the basis of the RTL functional description, a signal transmission relationship between registers at logic cell connection information level, or a schematic (see FIG. 12) representing the net list with graphical symbols.

The translation section 62 converts the RTL functional description received from the input section 61, into a state transition diagram, a Boolean expression, a timing diagram, and memory specifications indicative of, for example, the type of memory, the number of bits, and the number of words.

The optimization section 63 has the following components: a state-transition-diagram optimization processing section 63a for optimizing the state transition diagram; a state machine generator 63b for generating a circuit (i.e., a state machine) corresponding to the optimized state transition diagram; a timing-diagram compiler 63c for compiling the timing diagram; a memory synthesis section 63d for memory synthesis according to the memory specifications; and an interface-section synthesis section 63e for synthesizing an interface section according to the compiled timing diagram and the synthesized memory. The optimization section 63 further has a logic optimization section 63f. If the input section 61 inputs an RTL functional description, then the logic optimization section 63f performs, based on the obtained state machine, the obtained Boolean expression, and the synthesized interface section, optimization of a logic in order to generate optimized logic cell connection information. On the other hand, if the input section 61 inputs either a net list or a schematic, then the logic optimization section 63f performs optimization of a logic for the net list or the schematic to generate connection information about the optimized logic.

The output section 67 provides either a net list representing the FIG. 3 logic circuit or a logic circuit diagram (a schematic) representing the net list.

The present invention is characterized by the cell mapping section 64 shown in FIG. 9. Cell mapping is carried out by the cell mapping section 64. Such cell mapping, which is an algorithm for generating the FIG. 3 SIC by logic synthesis according to cell connection information generated by the logic optimization section 63f, will now be described by reference to FIG. 13. Note that FIG. 13 mainly illustrates features of the present invention.

At step S1, a functional design process is performed using HDL descriptions. At step S2, the HDL descriptions are inputted, based on which a high-voltage (3V) logic cell library (3Vlib) is chosen from among logic cell libraries shown in TABLE of FIG. 14, and CC mapping is executed by this 3Vlib.

At step S3, the signal propagation delay of a CC thus mapped is calculated. The result of the calculation is then checked if it exceeds a design delay upper limit (DDUL). If the result exceeds the DDUL, then the HDL description, which was inputted at step S4, is modified or, a functional design process is performed again to create new HDL descriptions. For example, (a) if in a partial circuit shown in FIG. 15 a CC f lies between r1 and r2, (b) if the function of the CC f is comprised of a function A and a function B, and (c) if the signal propagation delay of the CC f exceeds the DDUL, then (i) the CC f is divided into a CC f1 and a CC f2 so that the CC f1 is assigned the function A and the CC f2 is assigned the function B and (ii) the FIG. 17 functional description is modified to the FIG. 18 description wherein another register is connected between the CC f1 and the CC f2, thereby increasing the number of registers to three (r1, r2, r3).

At steps S5-S9 (the first processing), every CC having a signal propagation delay below the DDUL is mapped into a CC of the first type driven by the low-voltage (2V) source 16. Conversely, for the case of a CC having a signal propagation delay above the DDUL, its front section is mapped into a CC of the second type driven by the high-voltage (3V) source 15 while the rear section is mapped into a CC of the first type driven by the low-voltage (2V) source 16.

The first processing (steps S5-S9) is carried out as follows. At step S5, all CCs are mapped into low-voltage (2V) CCs of the first type, in other words mapping is carried out by 2Vlib. At step S6, the signal propagation delay of each CC thus mapped is calculated for every signal propagation path. If the result is determined to exceed the DDUL, then every CC with a delay in excess of the DDUL is extracted at step 7. After such CC extraction, steps S8 and S9 are performed. Suppose that such an extracted CC is formed of m combinational circuit elements (CCEs). Then, at step SB, the nth (n=1) of the m CCEs is subjected to remapping by a high-voltage (3V) CC of the second type, or by 3Vlib. At step S9, the signal propagation delay of the CC thus remapped is compared with the DDUL. If the result of the comparison operation exceeds the DDUL, then the next CCE located next, in the direction of signal propagation, is subjected to remapping by a high-voltage (3V) CC of the second type. Such operation is repeated until the signal propagation delay of the post-remapping CC becomes lower than the DDUL.

At steps S10-S12 (the second processing), the following are performed. Step S10 checks for the presence of a mixture of two different CCs, a 2VCC and a 3VCC, so arranged that the output of the 2VCC serves as input to the 3VCC. Every 2VCC, which is found by step S10 to form such a mixture with a 3VCC, is extracted at step S11. At step S12, remapping is performed in order that the extracted 2VCC of the first type is replaced with a 3Vlib which is a CC of the second type. After such remapping, the SIC logic synthesis algorithm returns to step S10, and steps S11 and S12 are repeated. The reasons for this is that remapping at step S12 may result in producing a new mixture.

The voltage type of CCs on the input and output sides of a register has already been determined by the foregoing logic synthesis, and the following are performed at steps S13-S15 (the third processing). At step S13, the individual registers are checked if they perform a level conversion in electric potential, in other words they are checked if they receive a low-level input (2V) and provide a high-level output (3V). A register that is found to perform no low-to-high level conversion, is mapped into a 2V FF circuit shown in FIG. 4 or into a 2V latch circuit shown in FIG. 7, at step S14. On the other hand, a register (a FF circuit or a latch circuit) that is found to perform a low-to-high level conversion, is mapped into a 2V/3V FF circuit shown in FIG. 5 or into a 2V/3V latch circuit shown in FIG. 8.

If every CC is subjected to mapping processing by 2VCC of the first type, this causes a CC 70 of FIG. 19(a) connected between registers to have a signal propagation delay in excess of the DDUL. Then a first CCE 70a and a second CCE 70b of the CC 70 at the front thereof (i.e., the first two CCEs on the signal propagation start side) are mapped into 3VCCs (see FIG. 19(b)), in the FIG. 13 algorithm. Thereafter, if the existence of a mixture of a 2VCC and a 3VCC, so arranged that the output of a CCE of the 2VCC serves as input to the 3VCC, is found, then the CCE (i.e., a CCE 71a of a 2VCC 71) is remapped into a 3VCCE (see FIG. 19(c)). Next, whether or not the remapping newly creates such a mixture is judged. For the case of FIG. 19(c), no mixture is found to exist. Accordingly, if a FF circuit is required to perform a level conversion from low-level input (2V) to high-level output (3V), the FF circuit is mapped to a 2V/3V FF circuit (hatched in FIG. 19(d)).

FIG. 19(d) shows a final SIC made in accordance with the present embodiment. One advantage of the present invention is that the number of CCEs to be mapped into a 3VCCE an be reduced therefore rendering the power consumption lower in relation to a case where all the CCEs of the CC 70 having a signal propagation delay in excess of the DDUL is subjected to mapping processing by 3VCCE.

EMBODIMENT 2

FIG. 20 illustrates a second embodiment of the present invention. The second embodiment employs a binary search method in which the boundary between the front and rear sections of a CC (i.e., the boundary between a CCE to be mapped into a 3VCC and a CCE to be mapped into a 2VCC) is defined.

At steps S18-S27 (the first processing), the boundary between the front and rear sections of a CC is defined by means of binary search. Whereas the front section is mapped into a CC of the second type driven by the high-voltage (3V) source 15, the rear section is mapped into a CC of the first type driven by the low-voltage (2V) source 16.

The first processing is described in detail. At step S18, every CC is subjected to mapping processing by 2VCC. At step S19, each CC thus mapped is checked if it has a signal propagation delay above the DDUL. Every CC with an excess delay is extracted at step S20 for further processing. More specifically, at step S21, a plurality of CCEs contained in a CC, which are subjected to mapping processing by 2VCC, are assigned their respective addresses wherein the first of the CCEs is assigned an address of ks and the last is assigned an address of ke. The initial setting of ks is ks=1. The initial setting of ke is ke=m where the number m is the number of CCEs forming the CC. Thereafter, step S22 determines whether ke-ks=1 holds. If ke-ks=1, then no dichotomy can be made and the second processing immediately starts. However, it is originally set such that ke-ks.noteq.1, and therefore, at step S23, k (the intermediate value) is found by the formula of k=(ks+ke)/2. At step S24, the first to kth CCEs are subjected to mapping processing by 3Vlib and the (k+1)th to mth CCEs are subjected to mapping processing by 2Vlib.

The signal propagation delay of the post-mapping CC is calculated. At step S25, the result of the calculation is compared with the DDUL. If the result exceeds the DDUL, ks (the first address) is set equal to k (the intermediate value) at step S26 in order that each of the (k+1)th to mth CCEs is further divided into two parts. On the other hand, if the result is less than the DDUL, ke (the last address) is set equal to k at step S27 in order that each of the first to kth CCEs is further divided into two parts. Thereafter, the program returns to step S22. It is judged, when ke-ks=1 at step S22, that the boundary between the front and rear sections of the CC is found. The program proceeds to the second processing.

The second processing and the third processing of the present embodiment are the same as in the first embodiment and therefore they are not described here.

For example, a case is explained in which a CC, comprised of a first half and a second half each containing ten stages of logic gates, has a critical path. In such a case, the first-half is remapped into 3Vlib and the second-half is remapped into 2Vlib. Whether SPD (signal propagation delay) is greater than DDUL (design delay upper limit), i.e., SPD>DDUL, is examined. If the result is YES, (a) the second-half of ten stages is further divided into two parts and (b) the five stages thereof, which are linked to the end, are remapped into 2Vlib and the remaining stages are remapped into 3Vlib. If the result is NO, (a) the first-half of ten stages is further divided into two parts, (b) the five stages thereof, which are linked to the starting point, are remapped into 3Vlib and the remaining stages are remapped into 2Vlib. Whether SPD is greater than DDUL is examined and the same processing as above is repeated. The use of binary search makes it possible to find a boundary by performing remapping processing several times, thereby achieving a fast processing rate.

EMBODIMENT 3

FIG. 21 shows a third embodiment of the present invention. A boundary between the front and rear sections of a CC is estimated in the present embodiment.

At step S1, a functional design process is performed using HDL descriptions. The signal propagation delay of a CC in the case of mapping by 3Vlib is estimated at step S2. The result of the estimation is compared with the DDUL at step S3. If the result exceeds the DDUL, either the HDL description is modified or a functional design process is performed over again to create a new HDL description at step S4 (see FIGS. 15 and 18).

At steps S30-S37 (the first processing), a boundary between the front and rear sections of a CC is roughly defined. The front section is mapped into a CC of the second type driven by the high-voltage source 15 and the rear section is mapped into a CC of the first type driven by the low-voltage source 16.

The first processing is explained in detail. Every CC is subjected to mapping processing by 2Vlib. At step S30, the signal propagation delay of each CC thus mapped is estimated. At step S31, the result of the estimation is compared with the DDUL. Each CC, found by the compare operation to have a signal propagation delay below the DDUL, is subjected to mapping processing by 2Vlib, at step S32.

All CCs, found by the compare operation to have a signal propagation delay above the DDUL, are extracted for further proceeding. At step S34, the result of the delay estimation is divided by the DDUL and the quotient is used to calculate p (the ratio of the presence of 3Vlib to the presence of 2Vlib). At step S35, p is multiplied by the number of logic gate (CCE) stages forming a CC, to calculate a mapping range of 3Vlib (i.e., a range of logic gates forming the front section). Thereafter, whether or not each logic gate falls in the 3Vlib mapping range is decided at step S36. A logic gate determined to fall in the range, is subjected to mapping processing by 3Vlib at step S37. On the other hand, a logic gate determined not to fall in the range, is subjected to mapping processing by 2Vlib at step S32.

The second processing and the third processing of the present embodiment are the same as in the first embodiment and therefore they are not described here.

Here suppose that the ratio of the delay of 3Vlib to the delay of 2Vlib is 1 to 1.8 and that the DDUL is 50 ns. For the case of a critical path having a delay of 90 ns, the entirety of the critical path becomes a 3Vlib mapping range. For the case of a critical path having a delay of 50 ns, there exists no 3Vlib mapping range so that all CCEs are subjected to mapping processing by 2Vlib. Further, for the case of a critical path having a delay of 60 ns, 1/4 of the critical path from its starting point is defined as a front section. This range is a range to be formed by 3Vlib. For the case of a critical path having a delay of 70 ns, 1/2 thereof from its starting point is defined as a front section range that is subjected to mapping processing by 3Vlib. For the case of a critical path having a delay of 80 ns, 3/4 thereof from its starting point is defined as a front section range that is subjected to mapping processing by 3Vlib.

In the present embodiment the boundary between a CCE that is subjected to mapping processing by 3Vlib and a CCE that is subjected to mapping processing by 2Vlib is roughly defined by estimation. This increases the rate of processing, although the accuracy of calculation for defining the boundary is not high. In usual logic synthesis, a remapping process may be performed on the basis of the result of a previously performed logic synthesis. In such a case, if an initial logic synthesis is carried out according to the present embodiment and a subsequent remapping process is carried out according to the first embodiment or according to the second embodiment, this arrangement not only improves the rate of logic synthesis processing but also increases the accuracy of boundary calculation. The number of CCES that are subjected to mapping processing by 3Vlib can be minimized and the power consumption can be lowered.

EMBODIMENT 4

FIG. 22 shows a fourth embodiment of the present invention. If it is anticipated that the mapping of a specified CCE or CCEs of a CC by 3Vlib produces a low-power SIC, such a 3Vlib mapping process is pre-performed in the present embodiment.

As can be seen from FIG. 22, new steps S40-S45 are added before the first processing (steps S5-S9). At step S40, whether the logic designer designates a specified CCE to be mapped by 3Vlib is determined. If the logic designer designates a CCE to be mapped by 3Vlib, it is arranged so that such mapping is designated by HDL at step S41. For example, a designation is made to the normal functional description of an adder which adds eight items of input data a-h in order that the last add element is subjected to mapping processing by 3Vlib (see FIGS. 23 and 24). This designation is shown as "//low.sub.-- power.sub.-- synthesis.sub.-- high-voltage" in FIG. 23.

Thereafter, a functional description is inputted. Step S42 checks for the presence or absence of such a designation. If no such designations are found, then all the CCEs are subjected to mapping processing by 2Vlib. On the other hand, if there is found a designation, then the designated CCE is subjected to mapping processing by 3Vlib.

Step S44 checks for the presence or absence of a mixture of a 2VCCE and a 3VCCE so arranged that the output of the 2VCCE serves as input to the 3VCCE. If the checking operation shows that there is such a 2VCCE/3VCCE mixture, then a level shifter, shown in FIGS. 6(a) and 6(b), is placed at the stage before the 3VCCE at step S45.

The first to third processing of the present invention are basically the same as the ones in the first embodiment (FIG. 13) and therefore they are not explained. If no mixtures are found at step S10, then step S50 checks for the presence or absence of a level shifter between a 3VCCE and another 3VCCE. When a level shifter is found to lie between 3VCCEs, the level shifter is removed at step S51. The reason for this is that it is anticipated that, when a 2VCCE is replaced by a 3VCCE by means of remapping in the second processing, a level shifter may be connected between that 3VCCE and another 3VCCE.

In the present embodiment, the logic synthesis method of the first embodiment of FIG. 13, which uses the FIG. 24 functional description with no designation of subjecting a specified CCE to mapping processing by 3Vlib, is constructed as follows. In an adder shown in FIG. 25(a) having seven add elements (arithmetic elements), the first four add elements (the hatched ones in FIG. 25(c)) are subjected to mapping processing by 3Vlib and each of eight registers before these four add elements are subjected to mapping processing by a 2V/3V FF circuit. However, in accordance with the present embodiment, only the last-stage add element (the hatched one in FIG. 25(b)) is subjected to mapping processing by 3Vlib and level shifters are provided at the stage before the last-stage add element. Therefore, the number of CCEs subjected to mapping processing by 3Vlib is further reduced thereby cutting down the power consumption.

EMBODIMENT 5

Referring now to FIGS. 26 and 28, a fifth embodiment of the present invention is illustrated. The present embodiment is characterized as follows. For the case of a CC having a signal propagation delay above the DDUL, it is arranged in the foregoing embodiment such that the front section of the CC is subjected to mapping processing by the second-type CC driven by the high-voltage (3V) source 15. In the present embodiment, however, what section is to be subjected to mapping processing is chosen according to the area or rate of processing.

FIG. 26 shows part of the FIG. 21 flowchart of the third embodiment and FIG. 27 shows the remaining part thereof. Step S35 of the third embodiment is changed to steps S60-S69 and steps S70 and S71 are added between steps S10 and S13.

As step S34, the ratio of 3Vlib to 2Vlib in a single CC, p, is calculated according to the signal propagation delay estimation result and the DDUL. At step S60, p is multiplied by the total number of gate stages of the CC, to calculate the number of gate stages subjected to mapping processing by 3Vlib (i.e., the range of mapping processing by 3Vlib). At step S61, the number of windows contained in the CC, n, is calculated in which the mapping processing range is used as a search range (window). From step S62 on, the n windows are evaluated. More specifically, at step S62, a variable, k, is set to its initial value (=0). At step S63, it is set such that k=k+1. At step S64, a CCE within the first window is set. At step S65, the area and delay of the CCE of the first window are evaluated. Thereafter, at step S66, k is compared with n. If k<n, the program returns to step 63 at which the total areas and delays of CCEs of the second to nth windows are evaluated. At step S67, of the n windows, one that has the smallest total CCE area or the shortest delay is chosen. If the selected window is not the first window, then a level shifter is placed at the stage before the selected window.

If step S10 no longer finds the foregoing 2V/3V mixture, then step S70 checks for the presence or absence of a level shifter connected between 3VCCEs. If a level shifter is found to exist, then the level shifter is removed at step S71 since such existence has resulted from the remapping processing in the second processing.

The present embodiment has the following advantages. In FIG. 28, of the three windows the third window of FIG. 28(c) is the smallest total add element area (count) and therefore this range is subjected to mapping processing by 3Vlib. This minimizes the number of add elements subjected to mapping by 3Vlib therefore further lowering the power consumption.

EMBODIMENT 6

Referring now to FIGS. 33 and 34, a sixth embodiment of the present invention is described. In the fourth embodiment, a CCE of an adder located at the last stage is subjected to mapping by 3Vlib. In the present embodiment, a CCE of a parallel multiplier of the carry save type located at the last stage is subjected to mapping by 3Vlib.

FIG. 33 shows a functional description including such a designation that a CCE of a parallel multiplier of the carry save type located at the last stage is subjected to mapping processing by 3Vlib. This functional description is fed into the input (read-in) section 61 of the logic synthesis system 60.

FIG. 34 shows a parallel multiplier of the carry save type made by logic synthesis using a functional description fed to the input section 61. This multiplier is formed such that AND circuits 90, half adders (HAs), and full adders (FAs) are arranged in an array fashion. An multi-bit adder 91 is placed at the lowermost stage in the array. The adder 91 is subjected to mapping processing by 3Vlib and the remaining components are subjected to mapping processing by 2Vlib. Sixteen level shifters 92 are placed at the stage before the adder 91. Each of the level shifters 92 receives from an adder located at its preceding stage a signal of 2V, converts it to a high level of 3V, and outputs it.

Most of the above-described multiplier may be formed of usual adders. However, the adder 91 must be a fast adder. The adder 91 may be implemented by, for example, a carry look ahead adder for high-speed processing. Since the adder 91 is of the high-voltage (3V) type, this makes it possible to generate circuits that are superior in circuit scale as well as in power consumption. Additionally, faster multipliers can be achieved.

The present embodiment is described in terms of parallel multipliers of the carry save type. The present invention provides an arithmetic circuit applicable in subtracters, dividers, accumulating adders, accumulating subtracters, accumulating multipliers, and accumulating divider.

Although the description has been made for the functional block A, the present invention may be applied to the remaining functional blocks (i.e., the functional blocks B-D). The present invention may be applicable between the functional blocks excluding the functional block E.

In accordance with the present logic synthesis method, part of CCEs of a CC with a critical path are of the high-voltage (3V) type and a level shifter is placed in a register located at the stage before such a critical path-containing CC. Like cases where in a critical path-containing CC only the critical path is driven by a high voltage, the need for determining where to place individual level shifters in a critical path-containing CC is eliminated. Additionally, the number of level shifters required can be reduced, which makes it possible to design SIC with ease. Further, the number of critical path-containing CCs is negligible in relation to the total number of CCs contained in an entire SIC and only part of CCEs of a critical path-containing CC are driven by the high-voltage source 15. Therefore, the increase in power consumption is held low and the remaining CCs without a critical path are driven by the low-voltage source 16, whereupon the entire SIC requires less power. Low power SICs can be realized.

The FIG. 3 SIC of the present invention is compared with the FIG. 30 SIC of a prior art technique. This conventional SIC is first described. The CCs 100, 102, 104, and 106 have their respective signal propagation delay times, i.e., 6 ns, 14.4 ns, and 8 ns. Suppose that the time interval (delay) between the application at the input terminal of a flip-flop of a clock signal and the appearance at the output terminal of the clock signal is 2 ns. Since the CC 104 has a delay time of 14.4 ns which is the greatest among the other CCs, the maximum operating frequency of the FIG. 30 SIC is approximately 60.98 MHz from the formula of 1000/(2+14.4).

The FIG. 3 SIC of the present invention is explained. The critical path-containing CC 5, the front and rear sections of which are of the 3V (high voltage) type and the 2V (low voltage) type, respectively, has a signal propagation delay equal to the DDUL (for example, 20 ns). The signal propagation delay of the CCs 1, 3, and 7 with no critical path increases with the logic cell delay because the supply voltage is reduced from 3V down to 2V. It is assumed that, in the case of the FIG. 3 SIC, the low-voltage source causes the cell delay to become 1.5 times longer than is the case with the high-voltage (3V) source. Of the CCs 1, 3, and 7 with no critical path the CC 3 has the greatest signal propagation delay, 18 ns.

In the present invention, two types of power sources, i.e., the low-voltage (2V) source 16 and the high-voltage (3V) source 15, are provided. As a result of such arrangement, the critical path-containing CC 5 has a delay of 18 ns (it is assumed that the low-voltage source 16 and the high-voltage source 15 cause the delay to become 1.25 times longer than is the case with the high-voltage source 15). The time interval (signal propagation delay) between the application at the input terminal of an FF circuit of a clock signal and the appearance at the output terminal of the clock signal (data), is 2 ns, and the delay of the CCs 3 and 5 is 18 ns. Therefore, the maximum operating frequency of the present SIC is 50 MHz.

1000/(2+18)=50 MHz.

If the maximum delay is below the DDUL, this causes no problems. Even when the CCs 1, 3, and 7 with no critical path are driven by the low-voltage source 16 while the critical path-containing CC 5 is driven by both the low-voltage source 16 and the high-voltage source 15, the design maximum operating frequency can be met.

FIG. 35(a) shows a delay distribution in a conventional 3V SIC in which the DDUL is 20 ns (i.e., the distribution of time intervals (time delays) between the time a clock signal is applied to an FF circuit and the time the data signal of the next FF is provided out of the FF circuit). In other words, FIG. 35(a) shows a distribution of signal propagation delay which is a sum of a register delay and a CC delay. FIG. 35(b) shows a delay distribution in an SIC in accordance with the present invention. If the supply voltage is changed from 3V down to 2V in the conventional SIC, this causes the maximum delay time to increase from 20 ns to 30 ns and the critical path delay time exceeds the DDUL of 20 ns. Conversely, in the SIC of the present invention, only critical path-containing CCs whose signal propagation delay exceed 20 ns when mapped by 2Vlib are driven by the high-voltage source 15 and the remaining other CCs containing no critical path are driven by the low-voltage source 16. The DDUL of 20 ns can be met (see FIG. 35(b)).

Next, a conventional SIC and an SIC of the present invention are compared in circuit scale. If (a) the circuit scale of the conventional SIC is S, (b) 10% of all FF circuits are provided with a level shifter, and (c) there is a 10% area increase due to the difference in circuit structure between the 2V/3V FF circuit of the present invention and the conventional FF circuit, then the circuit scale of the present SIC is found by the following formula.

S.times.0.8+S.times.0.18+S.times.1.1.times.0.02=S.times.1.002 The circuit scale increase is limited to an increase of 0.25%.

If 5% of all FF circuits are provided with a level shifter, then the circuit scale of the present SIC is given by the following formula.

S.times.0.8+S.times.0.19+S.times.1.1.times.0.01=S.times.1.001 The circuit scale increase is limited to an increase of 0.1%.

EMBODIMENT 7

Referring now to FIGS. 36 and 39, a seventh embodiment of the present invention is described. When using a combinational circuit of the high-voltage (3V) type in each of the foregoing embodiments, a register located at the stage before the combinational circuit is provided with a level shifter. The reason for this is to prevent a through current from flowing in the inverters 34a and 34b forming the output buffer 32 of the flip-flop circuit (register) 2 of FIG. 4. The reason of why such a through current flows is described in detail. FIG. 42 shows the internal structure of the inverter 34. Connected in series between Vo (high voltage of 3V) and VSS (ground) are a p channel transistor Tp and an n channel transistor Tn. The gates of Tp and Tn each receive a signal and a point where Tp and Tn are connected together serves as an output terminal of the inverter. The HIGH level of the input signal is 2V (the low-voltage source 16). When the input signal is HIGH, Tn turns on while Tp does not turn off completely. As a result, a through current flows from the high-voltage source 15 to the low-voltage source 16.

If the potential difference between the high-voltage source 15 and the low-voltage source 16 is below Vt (the threshold voltage of Tp), this ensures that Tp completely turns off, and no through current flows. If the potential S difference is set below Vt and if the high- and low-voltage sources 16 and 15 are a high-voltage (3V) source and a low-voltage (2.7V) source when, for example, Vt=0.5 V, such arrangement eliminates the need for providing level shifters. An example of this is shown below.

In a functional block shown in FIG. 36, the CCs 1, 3, 7 each are formed by a CC of the first type driven by the low-voltage (2.7V) source and the CC 5 is formed by a CC of the second type driven by the high-voltage (3V) source. The FF circuits 2, 4, 6 and 8 are of the 2.7V type. These FF circuits of the 2.7V type are identical in structure with the ones shown in FIG. 4 and are different from them only in power supply voltage.

FIG. 37 shows a logic synthesis method for the FIG. 36 functional block. Each CC is subjected to mapping processing by 2.7Vlib and at step S1 the signal propagation delay of the CC is estimated. Step S2 checks if the result of the estimation operation exceeds the DDUL. A CC with a signal propagation delay below the DDUL is subjected to mapping processing by 2.7Vlib (the first type) at step S3. On the other hand, a CC with a signal propagation delay above the DDUL is subjected to mapping processing by 3Vlib (the second type) at step S4. Thereafter, at step S5 each FF circuit is mapped into a 2.7Vlib FF circuit.

FIG. 38 shows a logic synthesis method. This method differs from the FIG. 37 logic synthesis method as follows. At-step S1', all CCs and FF circuits are mapped by 2Vlib. Step S2' checks each 2Vlib CC if it has a signal propagation delay in excess of the DDUL. At step S4', any CC judged to have an excess delay is remapped to a 2.7Vlib CC.

When the CC 70 connected between two registers has a signal propagation delay above the DDUL (see FIG. 39), the entirety of the CC 70 is subjected to mapping processing by 3Vlib, and an FF circuit located at the stage before the CC 70 is subjected to mapping processing by an FF circuit of the 3V type with no level shifter. In this case, it is arranged so that the CC 70 mapped by 3Vlib receives a low-voltage (2.7V) signal from the CCEs 71a, 71b. Since the CC 70 is able to operate normally even when it receives such a low-voltage signal, the CCEs 71a and 71b are not subjected to mapping processing by 3Vlib.

FIG. 40 shows a modification of the present embodiment. In the present embodiment, if there is a CC with a signal propagation delay in excess of the DDUL, then the entirety of that CC is mapped into a 3Vlib CC. In this modification embodiment, part of the CC is mapped into 3Vlib and the remaining is mapped into 2.7Vlib. What part is to be mapped into 3Vlib is determined the same way as the second embodiment (see FIGS. 26 and 27). In this modification, no level shifters are required. This eliminates steps S68 and S69 of FIG. 26 and steps S70 and S71 of FIG. 27.

Advantages of the present embodiment can be seen from FIG. 41. Here, DDUL=20 ns. If an SIC of the 3V type, the delay distribution of which is shown in FIG. 41(a), is changed to one of the 2.7V type, the maximum delay increases from 20 ns up to 24 ns (shown by broken line), exceeding the DDUL. In the delay distribution of a 2.7V/3V SIC of the present embodiment, however, the maximum delay time is 20 ns.

Claims

1. An arithmetic circuit comprising a plurality of arithmetic element rows each of which contains a given number of arithmetic elements wherein (A) arithmetic elements contained in the first of said plural arithmetic element rows each receive an external signal, (B) arithmetic elements contained in the remaining other arithmetic element rows each receive an output from arithmetic elements contained in its preceding arithmetic element row, and (C) arithmetic elements contained in the last of said plural arithmetic element rows each provide an arithmetic operation result,

wherein:

said arithmetic elements contained in the last arithmetic element row are driven by a high voltage from a high-voltage source;

said arithmetic elements contained in the remaining other arithmetic element rows except the last arithmetic element row are driven by a low voltage from a low-voltage source;

a level shifter driven by a high-voltage from said high-voltage source is placed between an arithmetic element of the last arithmetic element row and another of its preceding arithmetic element row so that a low-voltage output signal from said arithmetic element of said preceding arithmetic element row is converted into an output signal having the same voltage as a high voltage from said high-voltage source.

2. An arithmetic circuit according to claim 1 wherein said arithmetic circuit is an adder having a plurality of adder elements.

3. An arithmetic circuit according to claim 1 wherein said arithmetic circuit is a parallel multiplier of a carry save type which has a plurality of AND circuits and a plurality of adders in an array arrangement and which further has a multi-bit adder placed at the lowermost stage of said array.