High-speed multiplier for microcomputer used in digital signal processing system

BACKGROUND OF THE INVENTION
This invention relates to integrated semiconductor devices and systems, and more particularly to a high-speed, miniaturized, electronic digital signal processing system in single-chip microcomputer form.
A microprocessor device is a central processing unit or CPU for a digital processor which is usually contained in a single semiconductor integrated circuit or "chip" fabricated by "MOS/LSI" technology, as shown in U.S. Pat. No. 3,757,306 issued to Gary W. Boone and assigned to Texas Instruments. The Boone patent shows a single-chip 8-bit CPU including a parallel ALU, registers for data and addresses, an instruction register and a control decoder, all interconnected using the Von Neuman architecture and employing a bidirectional parallel bus for data, address and instructions. U.S. Pat. No. 4,074,351, issued to Gary W. Boone and Michael J. Cochran, assigned to Texas Instruments, shows a single-chip "microcomputer" type device which contains a 4-bit parallel ALU and its control circuitry, with on-chip ROM for program storage and on-chip RAM for data storage, constructed in the Harvard architecture. The term microprocessor usually refers to a device employing external memory for program and data storage, while the term microcomputer refers to a device with on-chip ROM and RAM for program and data storage; the terms are also used interchangably, however, and are not intended as restrictive as to this invention.
Subsequent to 1971 when U.S. Pat. Nos. 3,757,306 and 4,074,351 were originally filed, many improvements have been made in microprocessors and microcomputers to increase the speed and capability of these devices and reduce the cost of manufacture, providing more circuitry and functions in less space, i.e., smaller chip size. Improved VLSI semiconductor processing and photolithographic techniques allow narrower line widths and higher resolution, providing added circuit density and higher speed, but circuit and system improvements also contribute to the goals of increased performance with smaller chip size. Some of these improvements in microcomputers are disclosed in the following U.S. Patents, all assigned to Texas Instruments: U.S. Pat. Nos. 3,991,305 issued to Edward R. Caudel and Joseph H. Raymond Jr.; 4,156,927 issued to David J. McElroy and Graham S. Tubbs; 3.934,233 issued to R. J. Fisher and G. D. Rogers; 3,921,142 issued to J. D. Bryant and G. A. Hartsell; 3,900,722 issued to M. J. Cochran and C. P. Grant; 3,932,846 issued to C. W. Brixey et al; 3,939,335 issued to G. L. Brantingham, L. H. Phillips and L. T. Novak; 4,125,901 issued to S. P. Hamilton, L. L. Miles, et al; 4,158,432 issued to M. G. VanBavel; 3,757,308 and 3,984,816. The devices described in these patents have been of the Harvard architecture and of the 4-bit type, particularly adapted for calculator or controller applications.
Additional examples of microprocessor and microcomputer devices in the evolutation of this technology are described in publications. In Electronics, Sept. 25, 1972, pp. 31-32, a 4-bit P-channel MOS microcomputer with on-chip ROM and RAM is shown which is similar to U.S. Pat. No. 3,991,305. Two of the most widely used 8-bit microprocessors like that of U.S. Pat. No. 3,757,306 are described in Electronics, Apr. 18, 1974 at pp. 88-95 (the Motorola 6800) and pp. 95-100 (the Intel 8080). A microcomputer version of the 6800 is described in Electronics, Feb. 2, 1978 at pp. 95-103. Likewise, a single-chip microcomputer version of the 8080 is shown in Electronics, Nov. 25, 1976 at pp. 99-105. Another single-chip microcomputer, the Mostek 3872, is shown in Electronics, May 11, 1978, at p. 105-110 and an improved version of the 6800 is disclosed in ELectronics, Sept. 17, 1979 at pp. 122-125. Sixteen-bit microprocessors based on minicomputer instruction sets evolved such as the part number TMS9900 described in a book entitled "9900 Family Systems Design", published in 1978 by Texas Instruments Incorporated, P.O. Box 1443, M/S 6404, Houston, Tex. 77001, Library of Congress Catalog No. 78-058005. The 8086, a 16-bit microprocessor evolving from the 8080, is described in Electronics, Feb. 16, 1978, pp. 99-104, while a 16-bit microprocessor identified as the 68000 (based on the 6800) is described in Electronic Design, Sept. 1, 1978 at pp. 100-107, and in IEEE Computer, Vol. 12, No. 2, pp. 43-52 (1979).
These prior 8-bit and 16-bit microprocessors and microcomputers have been general-purpose processors of the Von Neuman architecture with multiplexed address/data busses, and usually have been microcoded as described in U.S. patent application Ser. No. 209,915, filed Nov. 24, 1980 by Guttag, McDonough and Laws (now U.S. Pat. No. 4,402,043), or Ser. No. 253,624, filed Apr. 13, 1981, by Hayn, McDonough and Bellay, both assigned to Texas Instruments, and at pp. 28-34, IEEE Spectrum, March 1979, by McKevitt and Bayliss, or Proceedings 11th Annual Microprogramming Workshop, December, 1979 By Stintter and Tredenick. Microcoding, originally described by Wilkes in 1951, employs a control ROM to store microinstruction sequences entered by instruction words; the programmer works in a higher level machine code, so the number of assembly language code statements is supposedly reduced, and thus programming cost is reduced.
In contrast, a special-purpose high-speed microcomputer device according to the embodiment of the invention described herein departs from these contemporary microprocessor devices in several major respects in order to achieve substantial speed and performance advantages. This device is a non-microcoded processor of modified Harvard architecture.
It is the principal object to this invention to provide improved features of a microcomputer device and system, particularly one adapted for real-time digital signal processing. Another object is to provide a high-speed microcomputer of enhanced capabilities.
SUMMARY OF THE INVENTION
In accordance with one embodiment, features of the invention are included in a system for real-time digital signal processing employing a single-chip microcomputer device having separate on-chip program ROM and data RAM, with separate address and data paths for program and data. An external program address bus allows off-chip program fetch in an expansion mode, with the opcode returned by an external data bus. A bus interchange module allows transfer between the separate internal program and data busses in special circumstances. The internal busses are 16-bit, while the ALU and accumulator are 32-bit. A multiplier circuit produces a single state 16.times.16 multiply function separate from the ALU, with 32-bit output to the ALU. One input to the ALU passes through a 0-to-15 bit shifter with sign extension.

BRIEF DESCRIPTION OF THE DRAWINGS
The novel features believed characteristic of the invention are set forth in the appended claims. The invention itself, however, as well as other features and advantages thereof, will be best understood by reference to the detailed description which follows, read in conjunction with the accompanying drawings, wherein;
FIG. 1 is an electrical diagram in block form of a microcomputer system employing features of the invention;
FIG. 2 is an electrical diagram in block form of an MOS/LSI microcomputer device (including a CPU or central processor unit) employed in the system of FIG. 1 and utilizing features of the invention;
FIGS. 3a-3mm are timing diagrams showing voltage or event vs. time in the operation of the microcomputer of FIG. 2;
FIGS. 4 and 4a are greatly enlarged plan views of a semiconductor chip containing the microcomputer of FIG. 2, showing the physical layout of the various parts of the device;
FIGS. 5a-5i are electrical schematic diagram of particular circuits in the microcomputer device of FIG. 2.

DETAILED DESCRIPTION OF SPECIFIC EMBODIMENT
MICROPROCESSOR SYSTEM
The microcomputer device to be described herein is primarily used for signal processing, but concepts thereof may be used in processor devices of various configurations, and these devices may be used in many different systems; in one embodiment the microcomputer is used in a system shown in generallized form in FIG. 1. The system may be, for example, a voice communication system, a speech analysis system, a small "personal" or "home" computer, a single-board general purpose microcomputer, a word processing system, a computer terminal having local processing capability with display and typewriter keyboard, or any one of many applications of various types. The system includes a single-chip MOS/LSI central processing unit or microcomputer 10 which will be described in detail, along with a program or data memory 11 and input/output or I/O devices 12. Usually the I/O devices 12 for the typical system include analog-to-digital and/or digital-to-analog converters, a modem, a keyboard, a CRT display, a disc drive, etc. Often the I/O 12 includes coupling to a general purpose processor; that is the microcomputer 10 is an attached processor in a larger system with interface via the I/O 12. The microcomputer 10, program data memory 11 and I/O 12 communicate with one another by two multibit, parallel address and data busses, D and RA, along with a control bus 13. The microcomputer 10 has suitable supply voltage and crystal-input terminals; for example, the device employs a single +5 V Vcc supply and ground or Vss, and a crystal is connected to terminals X1 and X2 of the device 10 to control certain system timing. The microcomputer 10 is a very high speed device with a crystal input of 20 MHZ, providing an instruction execution rate of five million per second, in one embodiment.
The microcomputer device 10 is a general purpose microcomputer specifically aimed at serving a large class of serial signal processing problems such as digital filtering, signal handling for telecommunications modems (modulation, demodulation), data compression for linear predictive code (LPC) speech signals, fast Fourier transforms, and in general for virtually all computation intensive analog system functions, including detection, signal generation, mixing, phase tracking, angle measurement, feedback control, clock recovery, correlation, convolution, etc. It is suitable for applications which have computational requirements similar to those for control and signal processing, such as coordinate transformation, solution of linear differential equations with constant coefficients, averaging, etc. The device 10 is usually interfaced via I/O 12 to a general purpose processor such as a 99000, an 8600 or a 68000, to construct processing systems as will be explained.
It is understood that, even though described in the context of a microcomputer in the preferred embodiment, with an on-chip program ROM 14 and data RAM 15, nevertheless some concepts of the invention may be used in a single-chip microprocessor with all off-chip program memory and/or data memory instead of the on-chip memory illustrated. Indeed, modes of operation are provided which disable the on-chip memory. Also, a microcomputer is shown having two separate external program address and data busses instead of the multiplexed, bidirectional busses which are now common, but some features herein disclosed are applicable where busses are multiplexed. The advantage of separating the busses and separating program and data memory space is speed.
In general terms, the system of FIG. 1 functions in the following manner: the microcomputer 10 fetches an instruction word internally by accessing the ROM 14 or externally by sending out an address on the ROM address bus RA to the memory 11 (and RCLK-on control bus 13). If external, the instruction word is received back via the data bus D from the addressed location in the memory 11. This instruction is executed in the next machine cycle (of length of 200 ns defined by a 20 MHz clock or crystal X1, X2) while a new instruction is being fetched; execution of an instruction may include accessing the on-chip RAM 15 for an operand, or writing a result into data RAM 15, and an arithmetic or logic operation in ALU.
In the example to be described in detail, a 12-bit instruction address applied internally to ROM 14 or externally to the RA bus directly addresses 2.sup.12 or 4K words of program instruction or constants in ROM 14 and memory 11. When reading from memory 11, a DEN- (data bus enable bar) command is asserted on control bus 13. It is also possible to write into the memory 11, and for this purpose a WE- (write enable bar) command is asserted by the device 10 on one of the control bus lines 13; the memory 11 may contain read/write memory devices in some or all of the address space, so the WE- command permits a write function.
The I/O devices 12 are addressed as ports; this interface to external devices 12 is accomplished using the address and data busses RA and D control bus 13, but the I/O devices 12 do not occupy locations in the logical address space like the memory 11. This is in contrast to conventional memory-mapped I/O.
Data input/output via I/O or peripherals 12 employs a 3-bit field from the bus RA to select one of eight 16-bit ports in peripheral circuitry 12. The selected 16-bit port is then accessed for read or write via the bus D. This operation uses one of the two instructions IN or OUT, on the control bus 13, WE- is active for write or OUT, or DEN- is active for read or IN. A ROM clock RCLK- is active on control bus 13 on every machine cycle except when either DEN- or WE- is active; that is, the memory 11 is activated by RCLK- for possible instruction word access from off-chip in each machine cycle, but if accessing peripheral 12 using DEN- or WE- then the RCLK- does not occur.
A reset signal RS- on the control bus 13 clears the program counter and address buss RA (resets to zero), sets the data bus D in a high impedance state, and the memory controls DEN-, WE- and RCLK- in an inactive (high) state. All address and temporary data registers within the microcomputer 10 are cleared by a reset routine in the ROM 14, but the internal RAM is not cleared. In this manner, the peripheral circuitry 12 (such as a main processor) can assert control, or initiate a start-up or power-on sequence.
An interrupt signal INT- on the control bus 13 causes the microcomputer 10 to halt execution (saving the current ROM address) and go to an interrupt vector address, unless interrupts are masked by the program.
The ME/SE- line in the control bus 13 defines the memory expansion mode or systems emulator mode for the microcomputer 10. When this pin is held high (at +Vcc), the microcomputer executes from on-chip ROM and off-chip memory 11, but when low (Vss) the chip is in the systems emulator mode and execution is only from the memory 11 which is PROM, EPROM or RAM so the program can be easily changed.
THE MICROCOMPUTER CHIP
The internal architecture of the microcomputer 10 is shown in a detailed block diagram in FIG. 2. This device is a single-chip semiconductor integrated circuit mounted in a standard dual-in-line package or a chip carrier. Sixteen pins or terminals of the package are needed for the 16-bit data bus D, twelve to sixteen are used for the address bus RA (depending upon memory size) and the remaining terminals are used for the power supply Vcc and Vss, the crystal X1, X2, and the control bus 13.
In addition to the program and data memory 14 and 15, the microcomputer 10 contains a central processing unit or CPU for the system of FIG. 1, and this CPU includes a 32-bit arithmetic logic unit or ALU, a 32-bit accumulator Acc to hold operands and results, multiplier M separate from the ALU, a shifter S which is one input to the ALU, status or flag decode SD, and an instruciton decoder ID1 which receives part of the current instruction word and generates the control bits for the CPU and data memory portions of the device 10.
The program memory 14 has associated with it a program counter PC to hold the instruction address used to access the ROM 14 or sent out on bus RA to the memory 11, an instruction register IR to receive the instruction word from ROM 14, a stack ST to save program memory addresses, and an instruction decoder ID2 which receives part of the current instruction word and generates control bits for the program memory portion of the microcomputer.
Associated with the data memory 15 are two auxiliary address registers AR0 and AR1 for the data memory 15, a page register ARP to select between the registers AR0 and AR1 as the data memory address, and a data page buffer DP to hold certain bits of the data memory address.
The CPU is oriented around two internal busses, a 12-bit program bus (P-Bus) and a 16-bit data bus (D-Bus). Program access and data access can thus occur simultaneously, and the address spaces are separate. The microcomputer is thus of the Harvard architecture, although a bus interchange module BIM permits loading the program counter PC from Acc, for example, or accessing ROM 14 for constants via P-Bus, BIM and D-Bus.
The two major requirements for a signal processing microcomputer are high speed arithmetic and flexibility. Performance is achieved by using separate, principally on-chip program and data memories 14 and 15, a large single accumulator Acc and a parallel multiplier M. A special purpose operation, data move, is defined within the data memory 15 which further enhances the performance in convolution operations. Flexibility has been achieved by defining an instruction set as will be described with reference to Table A, incorporating memory expansion and a single level of interrupt.
The device can be configured with, for example, less than 2K or 2.sup.11 words of on-chip program memory 14 and the architecture allows for memory expansion up to 4K or 2.sup.12 words by the addition of external program memory in the memory 11. In addition, a separate mode allows the device 10 to be configured as a system emulation device; in this "system emulator" mode, the entire 4K memory space is external and the ROM 14 is not used.
The CPU
The arithmetic logic unit or ALU consists of thirty-two parallel stages, each separate stage performing an arithmetic or logic function on its two input bits and producing a one-bit output and carry/borrow. The ALU has two 32-bit data inputs ALU-a and ALU-b, and a 32-bit data output ALU-o to accumulator Acc. The AlU-a input is always from the accumulator Acc and the ALU-b input is always either from the shifter S or from a 32-bit product register P in the multiplier M. The particular function performed on data passing through the ALU is defined by the current instruction word in IR which is applied by the program bus P-Bus to an instruction decoder ID1. The source of the ALU-b input is defined by an input select circuit ALU-s which selects from these two alternatives, based upon the contents of the current instruction word, i.e., the outputs #C of the decoder ID1. The shifter S receives a 16-bit input Si from D-Bus and produces a 32-bit output So which is the input Si shifted from zero to fifteen places to the left. Left-shifted data is zero-filled, i.e., all right-hand bit positions are filled with zeros when data is shifted out to the left. A unique feature is that the high-order bit is sign extended during shift operations. The ALU operates in twos-complement. The shifter S includes a shift control Sc loaded with a four-bit value from P-Bus via lines Sp so an arithmetic instruction can directly define the number of bits shifted in the path from D-Bus to the ALU-b input.
In this description, the LSB is considered to be on the right and the MSB on the left, so left-shift is toward more significant bits. Bit-0 is the MSB and bit-15 is the LSB. Data is always in signed 2's complement in this architecture.
The multiplier M is a 16.times.16 multiplier using carry feed-forward, constructed in partly dynamic and partly static logic, to implement Booth's algorithm. One input to the multiplier M is the T register which is a 16-bit register for temporary storage of the multiplicand received from D-Bus via lines Ti. The other 16-bit input is via lines Mi from the D-Bus; this multiplier input may be from the data memory 15 or may be a 13-bit multiply-immediate value derived directly from the instruction word (loaded right-justified and sign-extended).
The ALU always receives the contents of the accumulator Acc as its ALU-a input, and always stores its output in Acc, i.e., Acc is always the destination and the primary operand. The unit will add, subtract and perform the logic operations of And, Or and Exclusive Or. The logic operation results are between the lower half of Acc (bits 16-31) and a 16-bit value from the data memory 15. Due to passing the data memory value through the shifter S (with zero shift), the operand for the logical operation result of the MSBs (bits 0-15) is zero. The final 32-bit result reaching the accumulator is thus in two parts: Bits 0-15 will be Acc bits 0-15 Anded (or Or'ed, etc) with zero; bits 16-31 of the result will be Acc bits 16-31 Anded (etc.) with the data memory value. The accumulator Acc output, in addition to the 32-bit ALU-a input, includes high and low 16-bit outputs Acc-H (bits 0-15) and Acc-L (bits 16-31); separate instructions "store accumulator high" SACH and SACL "store accumulator low" are provided for storing high and low-order Acc bits in the data memory 15.
The status decoder SD monitors the Acc whenever an instruction which updates Acc is executed. Four bits of SD are OV, L, G and Z. Accumulator overflow (or underflow) is indicated by the OV bit, Acc contents less than zero is indicated by the L bit, Acc greater than zero indicated by the G bit, and Acc equal zero indicated by the Z bit. Upon interrupt the OV bit is saved in an overflow flag register, but the other bits are available only up to the time the next accumulator instruction is executed.
The accumulator overflow mode is a single-bit mode register OVM (included in SD), directly under program control, to allow for saturated results in signal processing computations. When the overflow mode OVM is reset, overflow results are loaded via ALU-o into the accumulator Acc from the ALU without modification. When the overflow mode is set, overflow results are set to the largest, or smallest, representable value of the ALU and loaded into the accumulator Acc. The largest or smallest value is determined by the sign of the overflow bit. This allows a saturated Acc result in signal processing applications, modeling the saturation process of analog signals.
A separate status bit in SD monitors the condition of the currently used auxiliary register AR0 or AR1 and detects the all-zero condition of the least significant nine bits of the current auxiliary register (i.e. loop counter portion). This bit is used for a branch instruction conditioned on non-zero for the auxiliary register (BARNZ), "branch on auxiliary register non-zero."
The input/output status bit (I/O ST-) is an external pin which is part of the control bus 13 and provides "branch on I/O zero" instruction (BIOZ) to interrogate the condition of peripheral circuits 12. A zero level on the I/O ST- pin will cause a branch when sampled by the BIOZ instruction.
The bus interchange module BIM exchanges the low-order twelve bits of the 16-bit value on the D-Bus with the twelve bits on the P-Bus. This operation is not available to the programmer as an instruction, but instead is needed as an inherent operation in instructions such as table look up (TBLR A).
Program Memory Addressing
The program memory 14 is a ROM which is partitioned to produce a 16-bit output to instruction register IR, and this ROM employs a decoder 14a which selects one 16-bit instruction word based on an 11-bit or 12-bit address on input lines 14b. In the example embodiment, the ROM 14 contains less than 2K words, so an 11-bit address can be used, but the on-chip program memory could be expanded to 4K with a 12-bit address. The circuit of the ROM 14 is expecially adapted for fast access as will be explained. The address input 14b is received from the program counter PC which is a 12-bit register containing the address of the instruction following the one being executed. That is, at the time when the control bits #C are valid at the outputs of the instruction decoders ID1 and ID2 for one instruction, PC contains the address of the next instruction; an address in PC goes into decoder 14a and the next instruction is read from ROM 14 into IR, and the program counter PC is incremented via PCinc in preparation for another instruction fetch. That is, PC is self incrementing under control of a #C control bit from ID2. The output PCo from the program counter PC is also applied via lines RApc and selector RAs (and output buffers not shown) to the external RA bus via output lines RAo and twelve output pins of the microcomputer device. The RA bus (RA0 through RA11) contains the PC output via RApc when the selector RAs is in one mode, or contains the input RAi when executing I/O instructions IN and OUT. Whenever the address in PC is above the highest address in ROM 14, off-chip program addressing to memory 11 is assumed; however, the device is designed to operate principally with the on-chip ROM, so for many uses of the device off-chip fetches for program instructions would never be needed. The program counter PC may be loaded via input PCi and selector PCs from the P-Bus for branch or call instructions, or loaded from the accumulator Acc via Acc-L, D-Bus, BIM, P-Bus, PCp and PCi in a "call accumulator" CALLA instruction.
The register stack ST is used for saving the contents of PC during subroutine and interrupt calls. In the illustrated embodiment, the stack St contains four 12-bit registers constructed as a first-in, last-out push-down stack, although a larger or smaller number of registers could be used. The current contents of PC are saved by "pushing" onto the top-of-stack register TOS via lines PCst. Successive CALL instructions will keep pushing the current contents of PC onto TOS as the prior contents are shifted down, so up to four nested subroutines can be accomodated. A subroutine is terminated by execution of a return instruction RET which "pops" the stack, returning the contents of TOS to PC via lines PCt, selector PCs and input PCi, allowing the program to continue from the point it had reached prior to the last call or interrupt. When TOS is popped, the addresses in lower registers of ST move up one position. Each subroutine, initiated by a call instruction or an interrupt, must be terminated by a RET instruction.
In an example embodiment, the ROM 14 contains 1536 words, so the remainder of the 4K program address space, 2560 words, is off-chip in the memory 11. When the memory expansion control pin ME/SE- is high, at logic 1, the device interprets any program address in PC in the 0-to 1535 range as being on-chip address for the ROM 14, and any address in the 1536-4095 range as being an off-chip address so that the PC contents are sent out via RApc and RAo to the RA bus. An output strobe RCLK- generated by the decoder ID2 for every machine state enables the external memory 11 (except when IN or OUT instructions are being executed). When off-chip program memory 11 is accessed, the instruction word read the memory 11 is applied to the external bus D and thus to the internal P-Bus via input/output control DC and lines Dp; this is a 16-bit instruction and, like the output of ROM 14 via IR, it is loaded into decoders ID1 and ID2 for execution, or loaded into PC via PCp, or otherwise used just as an on-chip instruction fetch.
When the ME/SE- pin is at zero the device enters the system emulator mode wherein the entire 4K program address space is off-chip, so all PC addresses are applied to the RA bus via RApc and RAo. This mode is necessary when a user is developing systems or programs, prior to arriving at a final version of code for the ROM 14. That is, the microcomputer 10 can operate with no code permanently programmed into the ROM so that new programs (stored in RAM or EPROM in the memory 11) can be tested and debugged, then when the final code is extablished the chips 10 are produced in large volume with this code mask-programmed into the ROM 14.
In either mode, the first two program addresses 0000 and 0001 are used for the reset function. When the reset pin RS- is brought low, an address of all zeros is forced into the program counter PC, as will be explained. Also, the third address is reserved for an interrupt vector; when the INT-pin is brought low, an address of 0002 is forced into PC to begin an interrupt routine.
Data Memory Addressing
The data memory 15 in the example embodiment contains 144 16-bit words, and so an 8-bit address is needed on address input 15a to the RAM address decoder 15b. However, the RAM 15 may be constructed with up to 512 words, requiring a 9-bit address, so the addressing arrangement will be described in terms of address bits which are unused in some embodiments. Each 128 word block of the RAM 15 is considered to be a page, so a 7-bit address field in an instruction word from program memory 14 on P-Bus via input 15c is used to directly address up to 128 words of data memory 15. Two auxiliary registers AR0 and AR1 are employed in the example embodiment; however, up to eight of these 16-bit registers may be used, with the particular one currently being used as the source of the address for the RAM 15 being defined by the auxiliary register pointer ARP. With two registers AR0 and AR1, the pointer ARP is only one bit, but for an embodiment with eight auxiliary registers the pointer ARP is a 3-bit register. The 16-bit auxiliary registers AR0 and AR1 are under control of store, load or modify auxiliary register instructions SAR, LAR, and MAR as will be described. Nine-bit addresses from the low-order parts of the auxiliary registers may be applied to the address input 15a via selector 15d, lines 15e, selector 15f, and lines 15g. When one of the auxiliary registers is to be the source of the RAM address, the selector 15d uses the value on lines 15e as the address input 15a, whereas if the P-Bus is to be the source of the RAM address the selector 15d uses a 7-bit address from input 15c and a 1-bit (expandable to 3-bit or 4-bit) page address from the data page register DP. The selector 15f is controlled by the pointer ARP which is loaded from P-Bus as defined by an instruction. The auxiliary registers are used for indirect addressing wherein an instruction need not contain a complete address for RAM 15 but instead merely specifies that an auxiliary register is to be used for this address; such instructions can also specify increment or decrement for the auxiliary register selected, in which case the nine LSBs of AR0 or AR1 are changed by +1 or -1 via paths Inc. The auxiliary registers may be thus used as loop counters. The auxiliary registers are accessed by the D-Bus vis lines ARio so these registers may be used as miscellaneous working registers, or may be initially loaded to begin a loop count.
The data memory 15 is accessed via the D-Bus and an input/output circuit 15i, via lines 15j. Construction of the data memory is such that a data move wholly within the RAM 15 is permitted, according to an important feature of the microcomputer 10. Under instruction control, the data at one address can be moved to the next higher location in one machine cycle without using the ALU or D-Bus. Thus during an add, for example, the accessed data can be also moved to the next higher address.
Input/Output Functions
Input and output of data from the microcomputer chip 10 uses the data bus D and two of the lines of the control bus 13, these being data enable bar (DE-) and write enable bar (WE-). Two instructions, IN and OUT, are employed for the data input and output functions. The external data bus D is coupled to the internal data bus D-Bus by the input/output control and data buffers DC. The output buffers in D1 are tri-state, so the output to data bus D is always placed in a high impedence state except when IN or OUT is being executed; to this end, one of the controls #C from the instruction decode ID1 sets the output buffers in high impdence state whenever IN or OUT is not decoded. When the instruction IN is present, the control DC activates sixteen input buffers, so the external data bus D is coupled to the internal D-Bus via DC and lines Dd for data input. When the OUT instruction is decoded, a control #C from ID1 activates output buffers in DC so the internal D-Bus is coupled via Dd and DC to the external bus D.
Execution of an IN instruction will also generate a data enable DEN- strobe on line 13a from ID1, and will couple the D-Bus to the RAM 15 via 15a and 15j, so the data from external will be entered into on-chip data memory. The intended uses of the microcomputer as a signal processor require hundreds or thousands of accesses to RAM 15 for every off-chip reference. That is, a value will be fetched from off-chip then convolution or like operations performed using this new value and other data in the RAM 15, so thousands of instruction executions will transpire before another off-chip reference is needed. For this reason, the architecture favors internal data manipulation over off-chip data access.
Execution of an OUT instruction causes generation of an off-chip write enable WE- strobe on line 13b from ID1 and outputs data from RAM 15 via 15i and 15j, D-Bus, lines Dd and buffer DC to the external bus D. Referring to FIG. 1, this data may be written into one of the ports (selected by the 3-bit RAi value) in the peripherals 12.
Implicit in both the IN and OUT instructions is a 3-bit port address on lines RAi from ID1. This address is multiplexed onto the three LSBs (RA9-RA11) of the external address bus RA via selector RAs. Up to eight peripherals may thus be addressed. The remaining high order bits of the RA bus outputs are held at logic zero during these instructions.
THE INSTRUCTION SET
The microcomputer 10 of FIGS. 1 and 2 executes the instruction set of Table A. The Table shows in the first column the mneumonic or assembly language name of each instruction used in writing source code, followed in the second column by the object code in binary which is the form the code appears in the ROM 14 and in the instruction register IR. This binary code is decoded in ID1 and ID2 to generate all of the controls #C to execute the desired operation by accessing various busses and registers and setting the functions of the ALU. The Table also gives the number of cycles or machine states employed by the microcomputer in executing the instruction; note that all instructions except branches, calls, table look-up and input/output are executed in one state time. The microcomputer is not microcoded; the standard ALU instructions are executed in one state. The Table also shows the number of instruction words needed to execute each instruction; it is important to note that only branches and call direct require two instruction words. The right-hand column is a brief description of the operation for each instruction.
Most of the instructions of Table A show the low-order eight bits (bits 8-15) as "IAAAAAAA", which is the direct or indirect RAM 15 address for one operand. If the "I" bit, bit-8, is 0, the direct addressing mode is used, so the "A" field of the instruction word, bits 9-15, is employed as a direct address connected from IR through P-Bus, lines 15c and selector 15d to address input 15a. In this direct addressing mode, the auxiliary registers AR0-AR1 are not used.
For the instructions containing "IAAAAAA", the indirect addressing mode is specified by a 1 in the I field, bit-8, of these instructions. The input address on lines 15a for the RAM 15 will in this case be obtained from one of the auxiliary registers AR0 or AR1, and bit 15 will select which one. If bit-15 is 0, AR0 is used; if bit-15 is 1, AR1 is used. Thus bit-15 coupled from IR via P-Bus controls the selector 15f (and can be loaded into the ARP register). Since the number of auxiliary registers is expandable to eight, bits 13-15 of these indirect-address instructions are reserved for use with a 3-bit selector 15f and ARP register to define one-of-eight in the indirect addressing mode. Bit-10 to bit-12 are controls in indirect addressing: bit-10 causes the addressed auxiliary register to be incremented if 1, or no change if 0; bit-11 causes the addressed AR to be decremented if 1 or no change if 0; bit-12 if 0 causes bit-15 to be loaded into ARP after execution of the current instruction, or if 1 leaves the ARP unchanged.
This shift code SSSS used in many instructions of Table A is a four-bit field loaded into shift control Sc via Sp to define the number of spaces (zero to fifteen) that the data coming from the RAM 15 via D-Bus is left shifted as it passes through the shifter S on the way to the ALU-b input.
Although not material to the structure described herein, assembly language formats using the instruction set of Table A employ "A" to designate direct addressing and "@" to designate indirect. Thus, "ADD S,A" means and contents of memory location defined by the A field of the instruction word. "ADD A@" means add using contents of the data memory location addressed by the auxiliary register AR0 or AR1 selected by the existing contents of ARP. ADD S@+ means add using current contents of ARP to define AR then increment this auxiliary register for loop counting. ADD S@ is the same as previous except decrement by 1. ADD S@-, AR is same as previous except ARP is loaded with the value of bit-15 to define a new auxiliary register for subsequent operations.
The descriptions given in the right-hand column of Table A assume direct addressing. For indirect addressing, the above explanation applies.
The ADD instruction thus adds the 16-bit contents of RAM 15 (at location OAAAAAAA for direct, or the contents at the locations in RAM 15 selected by the chosen AR if indirect), shifted SSSS spaces left, to the 32-bit contents of the Acc, and stores the result in the Acc. ADDH does the same except only the high-order half of Acc is the source of one operand and destination of the result, and no shift is performed.
The subtract instructions SUB and SUBH subtract the addressed RAM 15 data from the accumulator and store the result in Acc, but are otherwise the same as add. The load instruction LAC loads Acc with the 16-bit data addressed by IAAAAAAA which is left-shifted by SSSS bits. Only ADD, SUB and LAC specify a shift.
There are four instructions associated with the auxiliary registers: SAR, LAR, LARK and MAR. Store auxiliary register SAR causes the contents of one of the auxiliary registers defined by RRR to be stored in the memory location IAAAAAAA; the load AR instruction LAR is the reverse of SAR. With the LARK instruction a constant K from IR (bits 8-15) is loaded into the AR defined by RRR; this 8-bit constant K is right-justified and MSBs set to zero in the 16-bit auxiliary register. The modify auxiliary instruction MAR causes one auxiliary register to be modified by bits-10 to 12 as above, but no add or memory 15 access is implemented. The MAR code is operative only in the indirect mode, I=1; in direct mode this instruction results in no-op.
The input/output instructions are written in assembly language as "IN PA, A" or "OUT PA, A", where PA is the 3-bit port address PPP output on bits 9-11 of the RA bus (generated from the decoder ID1 and coupled via lines RAi). IN enables DEN- and disables RCLK-, while OUT enables WE- and disables RCLK-. The peripheral devices 12 decode RA9-RA11 to select one of eight 16-bit ports or locations for read or write via the bus D. These instructions use two machine states so that the data input pins of bus D are free on the second state to allow external fetch of the next instruction from memory 11 instead of ROM 14.
The store accumulator instructions SACL and SACH, written as "SACL X,A" in assembly, cause the low or high order bits of Acc to be left-shifted XXX places and stored in the data memory 15 at the location defined direct or indirect by IAAAAAAA. The X field is not fully implemented in the example embodiment; for SACL only X=0 is allowed and for SACH only X=0, X=1 and X=5 are allowed. This shift is implemented in the accumulator circuitry itself rather than in the shifter S.
The arithmetic and logic instructions without shift code are ADDH, ADDS, SUBH, SUBS, SUBC, ZALH, ZALS, EXOR, AND, OR and LACK. These are all written as ADDH A, for example, in assembly language. ADDH causes the 16-bit data from the defined location in RAM 15 to be added to the high-order half of Acc and stored in the high-order half of Acc; actually the data from RAM 15 is left shifted sixteen bits in shifter S as it goes from D-Bus to the ALU-b input. The ADDS instruction means that the sign extension is suppressed in the shifter S; the data from RAM 15 is defined by A is treated as a 16-bit positive number instead of a signed 2's complement integer. SUBH and SUBS correspond to ADDH and ADDS except subtract is performed in the ALU.
The conditional subtract instruction SUBC is used in divide operations. The contents of the defined location in RAM 15 are subtracted from the contents of Acc and left-shifted fifteen bits, producing an ALU output ALU-o which, if equal to zero is left-shifted by one bit and a +1 is added, with the result stored in Acc. If the ALU output is not equal to zero then it is left-shifted by one-bit and stored in Acc (the +1 is not added). SUBC is a two-cycle instruction that assumes the accumulator is not used in the following instruction. If the following operation involves Acc then a NO OP instruction should be inserted after SUBC.
The "xero accumulator load high" instruction ZALH fetches the 16-bit word at the addressed location in the RAM and loads it into the high-order half of Acc (bits 0-15); the Acc has been zeroed; so the low-order bits 16-31 remain zero. The shifter S is in the data path from D-Bus via ALU to Acc, so a 16-bit shift is performed in ZALH to move the data to the high-order half. The ZALS instruction fetches a word from RAM and loads it into the low-order half of the zeroed Acc, with sign extension suppressed in the shifter S.
The logic operations EXOR, AND and OR are performed in 32-bit format, even though the operand fetched is sixteen bits. For EXOR, the high-order half of Acc is Exclusive Or'ed with zeros, concatenated with Exclusive Or of the fetched data with the low-order half of Acc, both halves of the result being stored in Acc. The same applies to OR and AND.
The load accumulator instruction LACK causes an 8-bit constant contained in the eight LSB's of the instruction word to be loaded into the eight LSB's of Acc, right justified; the upper twenty-four bits of Acc are zeroed. To accomplish this operation, the instruction word on P-Bus from IR (after ID1 and ID2 are loaded, of course), is coupled to the D-Bus by BIM, and thence to the ALU-b via shifter S (with no shift). The ALU performs "pass ALU-b" or add zeros to b, leaving the constant in Acc.
The data shift or data move instruction DSHT causes the contents of the defined location in the RAM 15 to be moved to the defined location plus one. This is accomplished internal to the RAM 15 without using the ALU or data bus D-Bus. The operation cannot cross a page boundry, however.
The "load T" instructions are used to set up multiply operations. LT causes the T register to be loaded from RAM 15 with the value defined by IAAAAAAA. The "load T with data move" instruction LTD employs an operation like DSHT in the RAM; the T register is loaded with the contents of the RAM 15 location defined by IAAAAAAA, then this same value is shifted to location IAAAAAAA+1, and also the contents of Acc is added in ALU to the contents of the P register with the result going to Acc. The LTA instruction is the same as LTD but without data move; the T register is loaded from RAM 15 and the P register is added to Acc, with result to Acc.
The multiply instruction MPY causes the 16-bit contents of T register to be multiplied in multiplier M (not using ALU) by the value from RAM 15 on the input Mi from D-Bus, with the 32-bit result going to the P register. The "mulitply constant" instruction MPYK causes the 16-bit contents of T register to be multiplied by a 13-bit constant C from the opcode in IR; the 32-bit result stays in P register. For MPYK, the constant is connected from IR to Mi via P-Bus, BIM and D-Bus.
The "load data page" instructions LDPK and LDP cause the data page register DP to be loaded with up to eight bits from the opcode itself or from the defined location in RAM 15. In the embodiment shown, the DP register is only one bit, but in other embodiments with a larger RAM 15 the DP register contains up to eight bits. The page address remains the same until a new load page instruction occurs.
The load status and store status instructions LST and SST are used in call subroutine or interrupts to save the contents of the status circuits SD, or restore status SD. These instructions are used instead of hard wired circuits for performing this function.
The disable and enable interrupt instructions DINT and EINT are used to mask or unmask the interrupt capability, i.e., these instructions reset or set a latch which determines whether or not the microcomputer 10 responds to the INT- pin.
An absolute value instruction ABS functions to assure that the accumulator contains only an absolute valve, i.e., if Acc is less than zero, the absolute value of Acc is loaded into Acc, but if Acc is greater than zero there is no change. Similarly, the zero accumulator instruction ZAC clears Acc.
The overflow mode instructions RAMV and SAMV cause the overflow mode latch OVM in the status decode SD to be set to 1 or reset to 0. When set, the ALU output is set to its maximum or minimum before loading into Acc upon overflow. This simulates the effect of saturating an amplifier in an analog circuit, and is useful in signal processing.
Three P register instructions PAC, HPAC and SPAC are used in manipulating data after a multiply MPY or MPYK. PAC loads the accumulator with the contents of the P register by passing the 32-bit data through the ALU without performing any operation to modify the data; actually the ALU-a input is zeroed and an Add is executed. The APAC instruction adds the contents of the P register to the contents of Acc, with the result going to Acc. Similarly, the SPAC subtracts the contents of P register from Acc, result to Acc.
The subroutine instructions are CALL, CALLA and RET. CALL is a two-word instruction; the first word is the opcode and the second is the absolute address of the first instruction in the subroutine. When CALL is decoded in ID2, PC is incremented to fetch the next instruction word which is the address, then the incremented contents of PC are pushed to stack ST. The subroutine ends in return RET which causes the address on TOS to be popped and loaded into PC. To save status, SST must be used before CALL, and LST inserted after RET. The CALLA instruction is unique for a Harvard architecture machine; this uses the contents of Acc as the subroutine address rather than using the next location addressed by PC+1. The low-order bits of Acc are transferred via Acc-L and BIM to the P-Bus and thus via PCp to the program counter PC. The incremented PC is saved in CALLA by pushing to ST just as in a CALL.
The table look up instructions TBLR and TBLW also employ the Acc as an address source. These instructions require three states to execute. The RAM 15 location defined by IAAAAAAA is transferred via D-Bus and BIM to P-Bus, and thus via PCp to PC, from whence this address is applied via RApc to the external RA bus, or to ROM 14.
The branch instructions all require two words, the first being the opcode and the second at PC+1 being the address. The low-order bits 8-15 of the opcodes are unused. Unconditional branch B loads the word at PC+1 into PC as the next address. BARNZ is conditional upon whether or not a loop counter, one of the auxiliary registers defined by ARP, is not-zero. BV causes a branch is the overflow bit OV in the status decode SD is a 1. BIOZ causes a branch is the IO bit from I/O ST- is a 1 in the status decoder SD. The six instructions BLZ, BLEZ, BGZ, BGEZ, BNZ and BZ are all dependent upon the defined condition in SD reflecting the condition in Acc.
SYSTEM TIMING
Referring to FIGS. 3a-3ii, the timing of the system of FIG. 1 and the CPU chip of FIG. 2 is illustrated in a sequence of voltage vs. time waveforms or event vs. time diagrams. The chip 10 includes a clock generator 17 which has two external pins X1 and X2 to which a crystal (or external generator) is connected. The basic crystal frequency is up to 20 MHz and is represented by a clock O of FIG. 3a. This clock .PHI. has a period of 50ns, minimum, and is used to generate four quarter-cycle clocks Q1, Q2, Q3 and Q4 seen in FIGS. 3b-3e, providing the basic internal timing for the microcomputer chip 10. A set of four quarter cycle clocks Q1 to Q4 defines one machine state time of 200 ns, minimum; the states are referred to as S0, S1, S2, in FIG. 3. The clock generator 17 produces an output CLKOUT, FIG. 3f, on one of the control bus lines 13. CLKOUT has the same period as Q1, but 50% duty cycle and beginning at the midpoint of Q1. This output is used for timing or synchronizing external elements of the system of FIG. 1.
Internally, the microcomputer 10 executes one instruction per state time for most types of instructions, so five million instructions per second are executed. Of course, some instructions such as input/output, branch, call or table look-up require two or three state times. Assuming a sequence of single-state instructions such as add, load, store, etc., a new address is loaded into PC during each Q3 as seen in FIG. 3g, then the ROM 14 is addressed during Q4 and Q1 so an instruction word output is produced from IR onto P-Bus starting in the next Q2 and continuing through Q3, as seen in FIG. 3h. The ROM 14 access time is thus about 100ns. If an external instruction fetch from memory 11 is used, the same access time applies. The instruction decoders ID1 and ID2 receive the instruction word from P-Bus during Q3 as seen in FIG. 3i, and most of the decoder outputs #C are valid during Q1, although some fast controls are available in Q4. For direct addressing of the RAM, the address on bit-9 to bit-15 of P-Bus is immediately gated into the RAM decoder 15b when P-Bus becomes valid, but in either direct or indirect the RAM address is valid by the beginning of Q3 as seen in FIG. 3j. For RAM read, the data output via 15j to D-Bus is valid on Q4, FIG. 3j, and this data passes through the shifter S, FIG. 3k, and is available as an ALU input during Q1, FIG. 3l. The ALU controls #C are valid in Q2 and the ALU output ALU-o is available during Q3. The accumulator Acc is loaded from ALU in Q4, FIG. 3m.
It is thus seen that an ADD instruction, for example, for which fetch began at Q3 of the SO state in FIGS. 3a-3m, will be completed, i.e., the result loaded into Acc, in Q4 of state S2. There is substantial overlap of instruction execution. A new instruction fetch begins during Q3 of each state time, so execution of two more instructions have begun before one is finished.
Not shown in FIGS. 3a-3m is the write-RAM function. The RAM 15 is always written into during Q2. Addressing the RAM is always during Q3, however. Thus, an instruction such as "store accumulator low" SACL is illustrated in FIGS. 3n and 3o. The RAM address is received from the instruction register via P-Bus on Q3 of S1 (assuming the SACL instruction was fetched beginning at Q3 of SO), and the write will not occur until Q2 of state S2. During the read slot, Q4 of S1, a refresh occurs for the addressed row of the RAM, then the same address stays until Q2 of state S2 for the write. The D-Bus is loaded from Acc during this same Q2, see FIG. 3n.
If the accumulator must perform the saturate function in the overflow mode, i.e., OVM set to 1, this will be performed after the load accumulator function of FIG. 3m. That is, for the ADD instruction of FIGS. 3a-3m, the Acc is saturated during Q1 if the next state S3, so that when the accumulator is accessed by the following instruction it will be available to load the D-Bus on Q2.
When an instruction uses the data move function within the RAM 15, the move operation occurs during Q1 as illustrated in FIG. 3o. Also, if the increment loop counter function is performed for the auxiliary registers AR0 and AR1, the increment (or decrement) is executed in Q1. The T register, auxiliary registers AR0 or AR1, ARP latch, DP register and stack ST registers are each loaded during Q2 of any state time if these functions are included in the current instruction.
The bus interchange module BIM always executes a transfer from D-Bus to P-Bus beginning in Q2, if this function is defined by the instruction. The transfer from P-Bus to D-Bus by BIM is begun during Q4. The D-Bus is precharged on Q3 of every cycle, so no data can carry over on D-Bus through Q3 of any state, nor can data be loaded to or from D-Bus during Q3.
The program counter PC is incremented by the PCinc path during Q3 of each state time. That is, the load PC function of FIG. 3g is the incremented value just generated.
Execution of a branch instruction is illustrated in FIGS. 3p-3r. If the instruction loaded into the decoders ID1 and ID2 during Q3 of state S0 is a branch, the status decode SD bits from the previous instruction are valid during Q1 of S1 so the decision of branch or not is made at this point. Meanwhile, of course, another instruction fetch has begun so if the branch condition is met the instruction delivered to P-Bus during Q2 of S1 is used as the next address; if the condition is not met, however, this instruction is discarded. Assuming the condition is met, the branch address is loaded from IR via P-Bus to PC during Q3 of S1, and the new instruction delivered to IR and P-Bus in Q2 of S2 then decoded and executed beginning at Q3 of S2, FIG. 3r.
A CALL instruction is executed in the same time sequence as a branch, seen in FIGS. 3p-3r, except no SD evaluation is needed, and PC+1 is pushed to stack ST during Q3 of S1.
A return instruction RET is a two cycle instruction as illustrated in FIGS. 3s-3u. If the instruction loaded into the decoders ID1 and ID2 during Q3 of state SO is RET, the instruction fetch which began with PC+1 and load PC in Q3 of S) is discarded and a pop stack function is performed in Q3 of S1 so the next instruction fetch is to the return address. The instruction fetched during Q4 of S1 is then decoded and executed beginning at Q3 of S2.
Input (or output) instructions are executed in two cycles as illustrated in FIGS. 3v-3x. Assume the opcode loaded into the decoder ID2 in Q3 of SO is IN. The instruction fetched beginning at Q3 of S0 is not used; execution is inhibited by the decode of IN. The contents of PC at Q3 of S1 are saved until Q3 of S2 for the next instruction fetch; that is, PC is recirculated back to PC by the increment path, but no increment is performed. The controls #C produced from decode of IN are available for two states. The RAM address is loaded from P-Bus on Q3 of S1, seen in FIG. 3y, and the data input reaches D-Bus on Q4 of S1 and is written into RAM 15 during Q2 of S2. The DEN- control is active from Q4 of S1 through Q2 of S2 for the IN function. An OUT instruction is executed like IN except the RAM 15 is read during Q4 of S1 and the WE-control is active instead of DEN-.
A table look up instruction is executed as shown in FIGS. 3aa-3cc. The TBLR opcode is decoded beginning at Q3 of SO and causes the Acc to be loaded via D-Bus to BIM in Q2 of S1, then PC is loaded via P-Bus from BIM in Q3 of S1 so the content of Acc is used as the next instruction fetch address. Meanwhile, execution of the instruction fetched beginning at Q3 of SO is inhibited by preventing a ROM read control #RR from loading IR with the ROM 14 output, at Q2 of S1. The incremented contents of PC from Q3 of SO are pushed to ST during Q3 of S1, then popped at Q3 of S2 as the next instruction address. The data fetched from ROM 14 (or memory 11) using the address from Acc during Q4/S1 to Q1/S2 is loaded onto P-Bus during Q2 of S2 where it remains until Q4 of S2 at which time the BIM accepts the data from P-Bus and then transfers it to D-Bus on Q2 of S3, the next state. The destination address for RAM 15 loaded into decoder 15b from P-Bus by Q3 of S1 and remains for two states, so the RAM write occurring at Q2 of S3 will use the RAM address defined in the original TBLR opcode.
One of the problems inherent in manufacturing microcomputer devices is that of testing the parts to determine whether or not all of the elements are functional. In many microcomputers, the instruction words read from the internal ROM are not available on external busses and so the ROM cannot be checked in any way other than by executing all possible functions, which can be lengthy. The device of FIG. 2 allows the ROM 14 to be read out one word at a time using the interchange module as illustrated in FIGS. 3ee-3hh. A test mode, not part of the instruction set of Table A, is entered by holding the I/O ST- pin at above Vdd, for example 10 V, and holding RS-low, producing an input to the decoders ID1 and ID2 causing a ROM output function in which the ROM 14 is accessed every cycle and PC incremented as seen in FIG. 3ee. The P-Bus receives the ROM output, FIG. 3ff, but the opcodes are not loaded into the decoders ID1, ID2. Instead, the BIM accepts the opcodes from P-Bus on Q4 of each cycle and transfers to D-Bus on the next Q2, as seen in FIG. 3hh.
The Chip Layout
In FIG. 4, the microcomputer 10 of FIGS. 1 and 2 is illustrated in chip layout form. This is a top view of an MOS/LSI chip which is about 150 mils on a side. A major part of the area of the chip 10 is occupied by the memory including the ROM 14 and RAM 15 with their address decoders, and by the 16.times.16 multiplier M. The ROM 14 has associated with it an X address decoder 14X and a separate Y address decoder 14y for instruction word output; twelve address bits are used to define one of up to 4096 16-bit words in the ROM 14, although in this example only 1536 are on-chip.
The RAM 15 has an X address decoder 15b-x which selects 1-of-72 row lines, and a Y address decoder 15b-y and sense amplifiers 15s which select 1-of-2 column lines, so only eight bits are needed for the RAM select in this embodiment (eight bits could accomodate a 256 byte RAM).
The busses RA and D have twelve or sixteen bonding pads on the chip (total of twenty-eight) for connection to external, and the areas of the chip around these bonding pads seen in FIG. 4 are occupied by the buffers used for the ports. It will be noted that the RA bus is only used for output, so only output buffers are needed for this port, while the D-Bus requires tri-state output buffers as well as input buffers.
The multiplier M, shifter S, ALU, accumulator Acc and auxiliary registers AR0, AR1 on the chip 10 of FIG. 4 comprise a 32-bit wide "strip" which is an array of rows (parallel to control lines #C) and columns (parallel to metal bus lines such as D-Bus and P-Bus and ALU and register bits) containing all of the 16-bit and 32-bit registers, ALU bits, and the like circuitry associated with the D-Bus and related control lines #C. As set forth in U.S. Pat. No. 4,402,044, issued to McDonough and Guttag, assigned to Texas Instruments, an important feature is that the 32-bit ALU and its associated 32-bit Acc registers, the two 16-bit AR registers, the shifter S, and the bus interchange BIM as described above are laid out on the MOS/LSI chip 10 in a regular strip pattern as seen in FIG. 4. Other circuitry connected to the D-Bus and the ALU-b input and having controls #C shared by sixteen or thirty-two bits may also be in the strip, such as AR and BIM. The Acc and the ALU each contain thirty-two bits or stages which are laid out in a regular pattern like memory cells, the bits arrayed and aligned horizontally and vertically as seen in FIG. 4. D-Bus and P-Bus of FIG. 2 are each sixteen parallel metal strips on top of the cells of the ALU and registers, and all of the dozens of control lines #C are horizontal polysilicon lines typically used as the control gates for transistors in the ALU and its associated registers and like circuitry. This layout arrangement is advantageous because the multiplier ALU and registers, and perpendicular control lines #C and metal busses fit in an array with virtually none of the wasted space used merely for routing conductors in conventional construction of microprocessors. Metal bus lines such as P-Bus and D-Bus and control lines #C are in large part routed over functional regions or cells of the chip in the strip rather than over unused silicon, and many 90 degree turns are produced inherently at functional cells rather than in conductor routing. In the prior devices, the controls, the registers, the ALU, etc. where islands of circuitry connected by busses or conductors. The enlarged view of FIG. 4a shows a small part of the strip, two-bits wide, illustrating the metal bus lines and the polysilicon control lines #C for an N-channel silicon gate MOS device made generally by a single-level polysilicon process according to U.S. Pat. No. 4,055,444, assigned to Texas Instruments. Various contacts (not shown) would be made within each register bit or stage from metal to silicon or to polysilicon. It is significant to note that many of the connecting lines from registers to busses illustrated in FIG. 2 are not physically lines or elongated conductors at all but instead are merely metal-to-silicon or metal-to-poly contact areas along the metal bus lines of FIG. 4a. That is, routing of 16-bit or 32-bit sets of parallel conductors is minimized by the strip feature, and the size of the chip 10 is reduced. All busses are not needed in any one part of the strip, and thus the pitch or maximum width of cells of the strip is minimized since the the metal-to-metal spacing is a critical limiting factor in bit density.
The internal program of the microcomputer 10 may be modified at the gate level mask in making the chip. The macro code or program in the ROM 14 is defined by a single mask in the manufacturing process as set forth for example in U.S. Pat. Nos. 3,541,543, 4,208,726 or 4,230,504, assigned to Texas Instruments. By rewriting this user or macrocode, keeping the instruction set defined by ID1 and ID2 the same, a wide variety of different functions and operations are available.
Arithmetic Logic Unit
A detailed schematic diagram of one bit of the 32-bit ALU is shown in FIG. 5a. The ALU operates under control of six of the #C commands from the instruction decode ID1, these commands being labelled #AUM0-#AUM4 (valid on Q2) and #AUAB (valid on Q1). The ALU-a input, inverted, is on line AUa and the ALU-b input inverted, is on line AUb, both being valid on Q1, one from Acc and the other from the shifter S or P register. The ALU output is at line AUc, valid on Q4, representing one of the inverted 32-bit parallel output ALU-o to Acc. Table B shows the function produced by operation of the ALU for various combinations of the six #C commands. This ALU is generally the same as U.S. Pat., issued to Karl M. Guttag, assigned to Texas Instruments. Propagate and generate nodes AUp and AUg are precharged on Q1 and conditionally discharged by transistors AUd controlled by the ALU-a input, transistor AUe controlled by the ALU-b input and its complement, and transistors AUf controlled by the #AUM0-#AUM3 commands, according to the functions of Table B. A carry-out node AUh and a carry-in node AUi for each bit are coupled by a propagate-carry transistor AUj controlled by a line AUk which is the propagate node AUp inverted. The carry-out node AUh is precharged on Q1 and conditionally discharged via transistor AUm which is controlled by a NOR gate having the generate node AUg as one input and the absolute value command #AUAB as the other, so if #AUAB is 1 the transistor AUm is off and carry-out bar is always 1, meaning no carry or absolute value. If #AUAB is 0, the generate signal on AUg controls. The inverted propagate signal on AUk is one input to an Exclusive Nor circuit AUn with static load AUq; the inverted carry-in bar of line AUi is the other input to the Exclusive NOR, resulting in an output AUr which inverted is the ALU output AUc. The carry-in bar node AUi is made unconditionally 0 when control #AUM4 is high for logic functions OR, AND and EXOR, so this input to circuit AUn is unconditionally 1, but for ADD, SUB, etc., the control #AUM4 is 0 and the carry-in from the node AUu of the next low-order bit of the ALU controls.
The Shifter S
Referring to FIG. 5b, the shifter S includes a 16-bit input Si, a shift matrix Sm, a shift controller Sc, and a 32-bit output So going to the ALU-b input. The input Si is connected to receive the D-Bus at all times and to drive lines Sf in the matrix Sm through high level buffers. If no shift is to be performed, a line Sg is high, turning on all sixteen of the transistors Sh for this line, so the 16-bit data on lines Sf will appear on the sixteen right-most output lines So via diagonal lines Sj. All of the lines Sf are precharged on Q3 via thirty-two transistors Sk then conditionally discharged by the input Si. The sign bit is extended by detecting the MSB bit-0 of the input Si by the line Sm. A gate Sn also receives a #NEX not extend command from ID1 (one of the controls #C) to kill the sign extension for certain instructions of Table A. Based on the incoming sign bit Sm and #NEX, the gate Sn generates an extend command on line Sq to transistors Sq'. The transistors Sq' in series with lines Sr conditionally discharge the nodes Ss on lines Sf through transistors St. The control Sc is a 1-of-16 decoder or selector which receives the bits 4-7 of the instruction word from the P-Bus on 4-bit input Sp during Q3 and activates one of the sixteen lines Su; the lines Su are precharged in Q3 via transistors Sup and conditionally discharged during Q4 via transistors Sud and Sc'. The controls for the shifter S consist of the 4-bit value on Sp (the SSSS field of the ADD instruction, for example) defining the number of positions of left shift, and controls on lines #C for negating sign extension, etc. Since the data is usually in two's complement, the sign bit is extended to all bits to the left of the most significant data bit. The sign bit is 0 for positive and 1 for negative. If the shift is to be seven bits, for example, the seventh line Su stays high on Q4 and all others go low. This turns on all transistors Sh and St in the seventh row and all other transistors Sh and St are off. The 16-bit data coming in on lines Si is thus moved via transistors Sh and lines Sj to a position on lines So seven bits to the left of the zero shift (right-most) position, and zero-filled to the right due to the prcharge Sk. To the left, the sign bit will stay 0 if the bit-0 is low, but if bit-0 is 1 then Sq is high, transistors Sq are on, allowing all bits to the left to discharge.
Bus Interchange Module
The bus interchange module BIM, shown in detail in FIG. 5c, consists of sixteen identical stages, only one of which is illustrated. Each stage has two clocked inverters Ia, with no feedback loop since data is not held in BIM longer than about half a state time. Input node Ib is connected to the respective bit of P-Bus via one of sixteen transistors Ic driven by a control bit #BIFP valid on Q4. The D-Bus is connected to the input node Ib via transistors Id driven by the control bit #BIFD (Bus Interchange From D) from decoder ID1 valid on Q2. The output node Ie is connected to the P-Bus by a push-pull stage including transistors If and Ig, and a transistor Ih driven by a control bit #TP, valid during Q2 and Q3. Likewise, output node Ie is coupled to the D-Bus via a push-pull stage having driver transistors Ii and Ij, and a transistor Ik driven by a control bit #BITD valid on Q2 and Q4. The transistors Ig and Ij are driven by node Im at the output of the first inverter Ia, providing a push-pull output. Data is transferred from D-Bus to nodes Ib, Im, Ie on Q2, and then from these nodes to P-Bus on Q4. Similarly, data is transferred from P-Bus to nodes Ib, Im, Ie on Q4, and then from these nodes to D-Bus on Q4 or the next Q2.
The Multiplier
Referring to FIG. 5d, a schematic representation of the multiplier M and its T and P registers is shown, and corresponding detailed circuit diagrams are shown in FIGS. 5e, 5f. The 16-bit output of the T register is applied to a set of eight Booth's decoder Mb which produce eight sets of outputs Mc, each set including five functions: shift or no shift, and add, subtract or zero. A set of eight banks of 17-bit static carry-feed-forward adders Ma-1 to Ma-8 receive the Mc inputs when the T register is loaded, and so a significant part of the multiplication function is initiated before the MPY instruction is executed. The adders Ma-1 to Ma-8 are static in that no clock Q1-Q4 is needed to cause them to operate. Each stage of each level or bank includes a control section Mm responsive to the decoder outputs Mc, and the control section feeds an adder Mn. Level Ma-2 uses half adders and levels Mc-3 to Mc-8 use full adders. The first level Mc-1 does not need an adder because there is no partial product from the prior stage, so it has only the control section Mm. When the MPY instruction is decoded, on Q4 the second operand is applied to the static adders Mn from D-Bus by 16-bit input Mi. As each level of the eight levels of adders Ma-1 to Ma-8 calculates the sum, the partial product is fed forward via lines Mf to the next higher level, except for the two LSBs of each level which are fed to the dynamic adders Md via lines Me. When the static adder array settles, the 17-bit output Mf from the level Ma-8 plus the seven lower level 2-bit LSB outputs Me, is applied to a carry-ripple adder MD (31-stages) to perform the final carry evaluation, producing a 31-bit product in two's complement notation. The 31-bits are sign extended to obtain a 32-bit product in the product register P.
Booth's 2-bits algorithm reduces the number of adder stages to about half the number otherwise required. When performing multiply in the classic pencil and paper method, the right or LS digit of one operand is multiplied by the other operand to produce a partial product, then the next digit is multiplied to produce another partial product which is shifted one digit with respect to the first. Booth's algorithm gave a method of multiplying in binary which allowed two bits to be treated each time, instead of one. Thus, level Ma-1 multiplies the two LSBs of T reg times all bits of D-Bus, producing a partial product Me and Mf. The second level Ma-2 multiplies the next two bits of T reg by D-Bus, adds the partial product Mf from Ma-1, and generates a new partial product Mf and two more bits Me because this operation shifts two bits each level.
In FIG. 5e, one of the eight decoders Mb is shown, along with two bits of the T register. The T register stage consists of two inverters Ia with a recirculate transistor Rc clocked in Q4. The stage is loaded via transistor Ta by a #LT command from ID1 occurring on Q2 during an LT instruction. The outputs of two stages of the T register and complements are applied by lines To and Tc to one Booth decoder Mb. The decoder consists of four logic circuits, each having a static load Ba, Bb, Bc or Bd and a pattern of transistors Be with the lines To and Tc applied to the gates. Two of the terms have 1 or 0 fixed in the gates by lines Bf. Outputs Mc-1 and Mc-2 represent no-shift and shift commands and come from the logic stages Bc and Bd. Outputs Mc-4 and Mc-5 are true and complement outputs from load Ba of the first of the logic circuits, and these represent add and subtract commands. The output Me-3 from Bb is the zero command.
The first level Ma-1 of the static adders is simpler than the higher levels in that only the D-Bus input Mi and the inputs Mc are involved, with no partial product. The stages of this first level are as seen in FIG. 5e, along with the seventeen stages of level Ma-2 and level Ma-3, assuming zeros for the inputs Mm-6 and Mm-5 from the prior stage. The control sections Mm are all the same on all levels. Note that no elements are clocked.
The decoders Mb and control sections Mm with controls Mc define the Booth's two-bits at a time algorithm which reduces circuitry and increases speed by a factor of two. When two bits are interrogated successively, the only operations required are add, subtract, do nothing or shift by one bit. Considering the input from T as one operand, and from D-Bus as the other, the following table describes the function
______________________________________ PartialTi + 1 Ti (Ti - 1) Function Product______________________________________0 0 (0) Do nothing K + O0 0 (1) Add D K + D0 1 (0) Add D K + D0 1 (1) Shift D & Add K + 2D1 0 (0) Shift D & Subtract K - 2D1 0 (1) Subtract D K - D1 1 (0) Subtract D K - D1 1 (1) Do nothing K + O______________________________________
An example of multiplication using Booth's two bit algorithm is as follows: ##EQU1##
In the control sections Mm the inputs Mi from the D-Bus are controlled by a transistor Mm-1 and control Mc-1, not shift. The Mi input for the adjacent bit is gated in by transistor Mm-2 and the Mc-2 shift command, providing the "2D" function as just described. The zero is provided by transistor Mm-3 and zero control Mc-3 which results in node Mm-4 being connected to Vcc (zero in two's complement). The carry-in from the prior stage is on line Mm-5, and the partial product from the prior stage is on line Mm-6. The add or subtract control is provided by transistors Mm-7 controlled by the Mc-4 and Mc-5 add and subtract commands. The full adder includes logic gate Mn-1 receiving the outputs of the control section, as well as gates Mn-2 and the exclusive Nor Mn-3, producing a sum on line Mn-2 and a carry on line Mn-5. Speed is increased by using carry feed forward instead of carry ripple on the same level. Level Ma-1 has no partial product or sum Mm-6 from the prior stage, nor carry-in Mn-5, so the adder is not needed, only the control, producing a sum (a difference) at node Mn-8 and no carry. The second level Ma-2 is a half adder since no carry feed forward is received from Ma-1.
One of the adder stages of the 31-stage ripple-through carry adder is shown in FIG. 5f, along with one stage of the P register. The adder stage receives two inputs Me, (or Mf), gated on Q1 or Q3 by transistors Md1. The six LSBs of adder Md have their inputs gated in on Q1 because the static array levels Ma-1, Ma-2 and Ma-3 will have settled and outputs Me will be valid at this point, so the add and ripple through in Md can begin, although the outputs Mf are not yet valid. Thus, the more significant bits are gated on Q3 at transistors Md1. A carry input Md2 from the next lower-significant stage is applied to one input of an exclusive NOR circuit Md3, and to a carry output gate Md4 which produces a carry output Md5 to the next higher stage. A propagate term is generated from the inputs Me and the carry-in by logic gate Md6, and a carry generate term by a logic gate Md7 with Md4. The sum output Md8 is connected by line Md9 to the input of the P register stage, gated by #LPR (load P Reg) from ID1 on Q4 by transistor Pa. The P register stage consists of pair of inverters Ia and recirculate transistor Rc gated on Q2. The output is applied to the ALU-b input on Q1 by gate Pb with #NRPR (not read P Reg) from ID1 as one input, along with an inverter Pc. Transistor Pd precharges the ALU-b input on Q4.
The timing of the multiplier operation is illustrated in FIGS. 3jj to 3mm. On Q2 of So, the T register is loaded and outputs Mc from the Booth's decoder become valid. The Mi inputs from D-Bus are valid at Q4 of S1, assuming the MPY instruction is valid in decoder ID1 at Q3 of S1. The lower bits of the dynamic adder Md are loaded with Me on Q1 of S2, via Md1, and the carry begins to ripple through the lower of the 31-bits, then this continues in Q3 of S2 through the output Mf of the upper levels, so P register is loaded on Q4 of S2 via Pa, where the data remains until loaded to ALU-b on Q1 of a succeeding cycle.
The RAM
The cell used in the RAM 15 is a pseudo-static 6-transistor cell as seen in FIG. 5g. This cell differs from the traditional 6-transistor static cell in that refresh transistors 15m are used in place of polysilicon resisters or depletion transistors used as load impedences. The implanted resistors or depletion devices are larger and interpose process complexities. The storage nodes 15n are connected through cross-coupled driver transistors 15p to ground; one transistor 15p is on and theother off, storing a 1 or 0. Read or write is through access transistors 15q to data and data bar lines 15r, with gates of the transistors 15q driven by a row address line 15s. Refresh is accomplished when the refresh line 15t is pulsed high allowing the node 15n which is at 1 to be charged back up to a level near Vdd, while the 0 node 15n will conduct the pulse of current to ground through the on transistor 15p. The row address on 15s is delayed slightly from the refresh line 15t so that both won't begin at the same time. In the timing sequence of the FIGS. 3a-3e, particularly FIGS. 3j and 3o, the cell of FIG. 5j is read in Q4 of any cycle, or written into on Q2.
Referring to FIG. 5h, several of the cells of FIG. 5g are shown in a column. The data and data bar lines 15r are precharged to Vdd-Vt on Q1 and Q3 by transistors 15u. The refresh address on lines 15t-0, 15t-1 and 15t-2, etc., occur in sequence, one-at-a-time, generated by a ring counter; for example, if the RAM 15 is partitioned in 64 rows, then a 64 bit ring counter generates one refresh address bit each state time, refreshing the entire array once each 64 states. The refresh pulse occurs on a line 15t during Q3, while transistors 15u precharge and equallize the data and data bar lines. A row address on a line 15s might begin to come up to 1 during the later part of Q3 since read access is in Q4, so the sizes of the transistors are such that nodes 15n will not be both forced to Vdd-Vt when transistors 15m and 15q are all turned on. The on transistor in the pair 15p will hold the 0 node lower than the 1 node. After the refresh pulse on 15t goes low, for a cell addressed for both read and refresh in Q3, the delayed Q3 address line stays high momentarily to assure that the zero-going line 15r will discharge at least slightly through 15q and 15p for the 0 side. Then a bistable sense circuit including cross-coupled transistors 15v is activated by transistor 15w having Q4 on its gate (delayed slightly to make sure Q3 has gone to zero). This flips the data and data bar lines to full logic level, after which the column access transistors 15y are activated for the addressed column and data can be read out onto the D-bus. Internal shift is implemented by lines 15x connecting nodes 15z to adjacent column lines 15r via transistors 15z activated by a RAM move command #RM from decoder ID1, occuring on Q4. The data is held until Q2 of the next cycle (after Q1 precharge of all data and data bar lines 15r) before being applied to the adjacent column for this move operation. Meanwhile, the row address may be incremented by 1; i.e., the next higher line 15t-1, etc., goes high so on Q2 the data is written into the next higher location.
The sixteen bits of the RAM 15 are arranged as seen in FIG. 5i, with column lines (data and data bar lines) 15r running vertical and row lines 15s horizontal. The RAM is only 32-columns wide, so the column select 15y is merely one-of-two, even or odd. There are in this embodiment 144 row lines 15s. The LSB of the address 15b to the RAM is the column address, even or odd. To implement the data move operation, on even columns the LSB of the address buffer is complemented, but for odd columns the LSB of the address buffer is complemented and also the row decoder output on line 15s is incremented.
TABLE A__________________________________________________________________________THE INSTRUCTION SETSource No. of No. ofCode Object Code-Binary Cycles Words DESCRIPTION__________________________________________________________________________ADD 0000SSSSIAAAAAAA 1 1 Add word at RAM address A (shifted S places to left) to Acc; Result to AccSUB 0001SSSSIAAAAAAA 1 1 Subtract word at address A (shifted S places to left) from ACC; Result to AccLAC 0010SSSSIAAAAAAA 1 1 Load Acc with word at address A (shifted S places to left)SAR 00110RRRIAAAAAAA 1 1 Store contents of Aux Reg number R at location defined by ALAR 00111RRRIAAAAAAA 1 1 Load Aux Reg R with value at location AIN 010 00PPPIAAAAAAA 2 1 Input value on external data bus, store in AOUT 01001PPPIAAAAAAA 2 1 Output value at address A to ext data busSACL 01010XXXIAAAAAAA 1 1 Store low order Acc bits in location A, shifted X places leftSACH 01011XXXIAAAAAAA 1 1 Store high order Acc bits in location A, shifted X places leftADDH 01100000IAAAAAAA 1 1 Add value at address A to high order Acc bits; result to Acc; no shiftADDS 01100001IAAAAAAA 1 1 Add Acc to value at address A sign extension suppressedSUBH 01100010IAAAAAAA 1 1 Subtract value at address A from high order Acc bits; result to Acc; no shiftSUBS 011000AAIAAAAAAA 1 1 Subtract with sign extension suppressedSUBC 01100100IAAAAAAA 2 1 Conditional subtract for divide; left shift ALU output and conditional+1ZALH 01100101IAAAAAAA 1 1 Zero Accumulator and Load High-order half of Acc with addressed dataZALS 01100110IAAAAAAA 1 1 Zero Accumulator and Load with sign Extension SuppressedTBLR 01100111IAAAAAAA 3 1 Table Read; read data from program memory using Acc or address; store in RAMMAR 01101000IAAAAAAA 1 1 Modify Auxiliary RegistersDSHT 01101001IAAAAAA A 1 1 Data Shift; value defined by A shifted to A+1LT 01101010IAAAAAAA 1 1 Load T Reg with value defined by ALTD 01101011IAAAAAAA 1 1 Load T Reg with value A; shift A to A+1; Acc+Preg AccLTA 01101100IAAAAAAA 1 1 Load T Reg with value defined by A; Acc+Preg AccMPY 01101101IAAAAAAA 1 1 Multiply T times value defined by A, result to P RegLDPK 01101110DDDDDDDD 1 1 Load page reg for data memory with 8-bit constant DLDP 01101111IAAAAAAA 1 1 Load DP reg with value whose address is at ALARK 01110RRRDDDDDDDD 1 1 Load Auxiliary Register R with 8-bit constant D; MSB's ZeroEXOR 01111000IAAAAAAA 1 1 Exclusive OR Acc with value defined by A; result to LSB; of Acc; zero.multidot.MSB'sAND 01111001IAAAAAAA 1 1 AND LSB's of Acc with value defined by A; result to LSB's of Acc; (zero).multidot.(MSB's)OR 01111010IAAAAAAA 1 1 OR LSB's of Acc with value defined by A; result to Acc; (zero)+(MSB's of Acc)LST 01111011IAAAAAAA 1 1 Load Status with 16-bit value found at location A in RAMSST 01111100IAAAAAAA 1 1 Store Status in location defined by 8-bit address A in RAMTBLW 01111101IAA AAAAA 3 1 Table Write; write the value at RAM address to program memory address in AccLACK 01111110DDDDDDDD 1 1 Load Accumulator with 8-bit constant from instruction wordNOOP 0111111110000000 1 1 No-operationDINT 0111111110000000 1 1 Disenable Interrupt-masks interrupt input INTEINT 0111111110000010 1 1 Enable Interrupt-unmasks interrupt input INTABS 0111111110001000 1 1 Absolute Value operation; if Acc 0, Acc Acc; else Acc AccZAC 0111111110001001 1 1 Clear Accumulator; zeros AccRAMV 0111111110001010 1 1 Reset Overflow ModeSAMV 0111111110 001011 1 1 Set Overflow ModeCALLA 0111111110001100 2 1 Call subroutine indirectRET 0111111110001101 2 1 Return from SubroutinePAC 0111111110001110 1 1 Load accumulator with contents of P RegAPAC 0111111110001111 1 1 Add accumulator to contents of P Reg; Result to AccSPAC 0111111110010000 1 1 Subtract contents of P Reg from Accumulator; Result to AccMPYK 100CCCCCCCCCCCCC 1 1 Multiply by constant CBARNZ 11110100XXXXXXXX 2 2 Branch if Loop Counter Not Zero, to location defined PC+1BV 111101 01XXXXXXXX 2 2 Branch if Overflow Bit is ST is 1BIOZ 11110110XXXXXXXX 2 2 Branch if IO Bit in ST (from IO pin) is 1CALL 11111000XXXXXXXX 2 2 Call SubroutineB 11111001XXXXXXXX 2 2 Unconditional Branch to location W at PC+1BLZ 11111010XXXXXXXX 2 2 Branch if Acc is less than zeroBLEZ 11111011XXXXXXXX 2 2 Branch if Acc is less than or equal to zeroBGZ 11111100XXXXXXXX 2 2 Branch if Acc is greater than zeroBGEZ 11111101XXXXXXXX 2 2 Branch if Acc is greater than or equal to zeroBNZ 11111110XXXXXXXX 2 2 Branch if Acc is not zeroBZ 11111111XXXXXXXX 2 2 Branch if Acc is equal to__________________________________________________________________________ zero
TABLE B__________________________________________________________________________ALU FUNCTIONS Control Code Propogate Generate #AUM0 #AUM1 #AUM2 #AUM3 #AUM4 #AUMB Node Node Output__________________________________________________________________________Add 0 1 1 0 0 0 A+B AB A+B+C.sub.inSubtract 1 0 0 1 0 0 A+B AB A+B+C.sub.inLoad Acc 0 1 0 1 1 0 B X B+1=BExclusive Or 1 0 0 1 1 0 A+B X A+B+1=A+BOr 1 0 0 1 1 0 AB (=A+B) X A+B+1=A+BAnd 0 1 1 1 1 0 A+AB X (A+AB)+1=A+B=ABAbs. Value 0 0 1 1 0 1 A 0 A+C.sub.in__________________________________________________________________________

Number	Name	Date
3691359	Dell et al.	Sep 1972
3795880	Singh et al.	Mar 1974
4153938	Ghest et al.	May 1979
4168530	Gajski et al.	Sep 1979

High-speed multiplier for microcomputer used in digital signal processing system

Information

Patent Number

Date Filed

Date Issued

Inventors

Original Assignees

Examiners

Agents

CPC

US Classifications

Field of Search

US

International Classifications

Abstract

Description

Claims

US Referenced Citations (4)

Non-Patent Literature Citations (3)

Entry
Frank S. Lee et al., "A High-Speed LSI GaAs 8.times.8 Bit Parallel Multiplier", IEEE Journal of Solid State Circuits, vol. SC-17, No. 4, Aug. '82.
Shlomo Waser, "High-Speed Monolithic Multipliers for Real-Time Digital Signal Processing", IEEE Computer, Oct. 1978, pp. 22-24.
J. P. Beraud et al., "Parallel Multiplier", IBM Technical Disclosure Bulletin, vol. 24, No. 11B, Apr. 1982.