Background of the Invention

This invention pertains to the field of digital computing and, more particularly, to a large scale digital computing system capable of performing high-speed addition without unduly increasing power consumption.

Addition is a fundamental operation in the arithmetic unit of a digital computing system because it is employed to perform, not only the addition function, but also the multiplication ^• function and the accumulation- function. In a conventional adder, the processing time is directly related to the number of bits in the numbers being processed because of the necessity for carries to propagate from stage to stage before the arithmetic result is generated. A number of schemes have been developed over the years, such as the so-called look ahead adder, to generate arithmetic results without the delays occasioned by carry propagation. These schemes require a great deal of extra electronic circuitry and, therefore, dis- sipate more power than a conventional adder.

Summary of the Invention

• According to the invention, sum and carry computations are processed separately during the execution of a plu- rality of arithmetic operations in tandemly-connected

arithmetic units so that carry propagation delay only occurs once, namely, at the end of the multi-operation computation. In the preferred embodiment, a digital computing system comprises first, second, third, fourth, fifth, and sixth multi-bit binary signal sources and first and second binary adders. The sum of the first and second signals represents a first binary number to be added. The sum of the third .and fourth signals represents a second binary ^' number to be added. The sum of the fifth and sixth signals represents a third binary number to be added. Each binary adder has a plurality of parallel stages equal in number to the bits of the signals. Each stage of each adder has a first full adder and a second full adder. Each full adder has an addend input, an augend input, a carry input, a sum output, and a carry output. In the first adder, the first source is connected to the addend input of the first full adder, the second.source is connected to the augend input of the first full adder, the third source is connected to the carry input of the first full adder, the sum output of the first full adder is connected to the addend input of the second full adder, the carry output of the first full adder is connected to the carry input of the second full adder of the next higher order - stage, and the fourth source is connected to the augend input of the second full adder.

A feature of the invention is implementation of the full adders with low-level, non-saturating, bipolar differential logic circuitry, which greatly reduces the power dissipation at high data processing ^' speeds. Separate processing of sum and carrier computations and use- of currently available low-level logic circuitry permits something on the order of one billion arithmetic operations per second to be carried out with an average power dis- sipation of only about one watt.

The foregoing measures increase the number of wire interconnections fourfold. A doubling of the wire interconnections results from the separate processing of carry and sum signals, and a doubling of the wire inter- connections results from the use of bipolar logic circuitry. To better organize and route this increased number of wire interconnections, a feature of the invention is the physical layout of the circuitry on an integrated circuit chip so the stages within an adder are contiguous to each other in bit order and/or the corresponding stages of different adders are aligned with each other.

Brief Description of the Drawings •

The features of a specific embodiment of the best mode contemplated of carrying out the invention are illustrated in the drawings, in which: FIG. 1 is a block diagram of one stage of a full adder that employs low-level, non-saturating, bipolar differential logic circuitry;

FIG. 2 is a schematic circuit diagram of the circuitry of FIG. 1; FIG. 3 is a schematic block diagram of three stages of an adder incorporating principles of the invention;

FIG. 4 is a schematic block diagram of a digital computing system incorporating a number of adders of the type disclosed in FIG. 3; FIG. 5 is a schematic block diagram illustrating on an integrated circuit chip the layout of stages within an adder and stages of adjacent adders.

Detailed Description of the Specific Embodiments

In FIG. 1, one stage of a full adder implemented with low-level, non-saturating, bipolar differential logic circuitry is shown. Such a full-adder stage is the basic building block from which the preferred embodiment of the invention is implemented. Single-ended input terminals X, Y, and C-j_ are connected to registers 10, 12, and 14, which have complementary outputs. One bit of a binary addend signal to be added is applied to ^• terminal X. The corresponding bit of a binary augend signal to be added is applied to terminal Y. The next lower order bit of a binary carry signal is applied to terminal C _j _. The complementary outputs of registers 10, 12, and 14 are connected to a sum circuit 16 and a carry circuit 18. Sum circuit 16 and carry circuit 18 perform the well-known logical functions required to generate a binary sum value and a binary carry value from the signals applied to-terminals X, Y, and-C_. These logical functions are described, for example, on page 161 of Digital Computer Fundamentals, Second Edition, by Thomas C. Bartee, McGraw- Hill Book Company, 1966. The complementary outputs from sum circuit 16 and carry circuit 18 are connected to registers 20 and 22, respectively, to transform them to single-ended form. Thus, register 20 has an output terminal S at which one bit of a binary sum signal appears, and register 22 has an output terminal C _D at which one bit of a binary carry signal appears.

FIG. 1 illustrates how the use of complementary signals doubles the number of wire interconnections vis- a-vis single-ended signals. In the preferred embodiment of the invention, data transmission from stage to stage of an adder and between successive adders is carried out by complementary signals within an integrated circuit chip. As taught in a co-pending, commonly-assigned application of Kelley and Stone entitled High-Speed

Digital Data Communication System (attorney docket PD:86402), the complementary signals are transformed to single-ended signals only to make inter-chip connections or at the end of the computation. In FIG. 2, sum circuit 16 is implemented by a hierarchy of transistors 24 to 33. The emitters of tran¬ sistors 24 and 25 are connected to the collector of a transistor 34. The emitter of transistor 34 is connected through a resistor 36 to a source of negative bias potential V _EE . The base of transistor 34 is connected to a conventional bias voltage source that compensates for changes in the characteristics of the circuitry with changing temperature as described in the above-identified co-pending application. The disclosure of said application is incorporated fully herein by reference. Input terminal C^ is connected to the base of transistor 24, and com¬ plementary input terminal C_ is connected to the base of transistor 25". Input terminal Y is connected to the bases of transistors 26 and 27. Complementary input terminal Y is connected to the bases of transistors 28 and 29. The emitters of transistors 26 and 28 are connected to the collector of transistor 24. The emitters of transistors 27 and 29 are connected to the collector of transistor 25. Input terminal X is connected to the bases of transistors 30 and 31. Complementary input terminal X is connected to the bases of transistors 32 and 33. The emitters of transistors 30 and 32 are con¬ nected to the collectors of transistors 27 and 28. The emitters of transistors 31 and 33 are connected to the collectors of transistors 26 and 29. The collectors of transistors 30 and 33 are connected to the base of an output transistor 36. The collectors of transistors 31 and 32 are connected to the base of an output transistor 38. The collectors of transistors 36 and 38 and the base and collector of a transistor 40 are connected to

ground. The emitter of transistor 40 is connected by a resistor 42 to the base of transistor 36. The emitter of transistor 40 is also connected by a resistor 44 to the base of transistor 38. The emitters of transistors 36 and 38 are connected to bias source V _E by resistors 46 and 48, respectively. Output terminal S is connected to the junction of the emitter of transistor 38 and resistor

48. Complementary output terminal S is connected to the junction of the emitter of transistor 36 and resistor 46. Carry circuit 18 is implemented by a hierarchy of transistors 48 to 55. The emitters of transistors 48 and 49 are connected to the collector of a transistor

56. The emitter of transistor 56 is connected through a resistor 58 to source V _EE . The base of transistor 56 is connected to a conventional bias voltage source that compensates for changes in the characteristics of the circuitry with changing temperature as described in the above-identified co-pending application. Input terminal

C_ is connected to the base of, transistor 48, and co - plementary input terminal Cj_ is connected to the base of transistor 49. Input terminal Y is connected to the bases of transistors 50 and 51. Complementary input terminal Y is connected to the bases of transistors 52 and 53. The emitters of transistors 50 and 52 are con- nected to the collector of transistor 48. The emitters of transistors 51 and 53 are connected to the collector of transistor 49. Input terminal X is connected to the base of transistor 54. Complementary input terminal X is connected to the base of transistor 55. The emitters of transistors 54 and 55 are connected to the collectors of transistors 51 and 52. The collectors of transistors 50 and 54 are connected to the base of an output transistor 56. The collectors of transistors 53 and 55 are connected to the base of an output transistor 58. The collectors of transistors 56 and 58 and the base and collector of a

transistor 60 are connected to ground. The emitter of transistor 60 is connected by a resistor 62 to the base of transistor 58. The emitter of transistor 60 is con¬ nected by a resistor 64 to the base of transistor 56. The emitters of transistors 56 and 58 are connected to bias source V by resistors 66 and 68, respectively. Output terminal C ₀ is connected to the junction of the emitter of transistor 58 and resistor 68. Complementary output terminal C _j _ is connected to the junction of the emitter of transistor 56 and resister 66.

Transistors 24 to 33 and 48 to 55 are preferably operated in a non-saturating linear mode with a low- level voltage swing of the order of 130-200 millivolts. Typically, these transistors could be FAST-Z FINELINE transistor type VN5H211, with the remaining transistors being type VN5H111.

In FIG. 3 is shown three stages of an adder incor¬ porating principles of the -invention. These stages, separated by vertical broken lines 70 to 76, process three bits, designated i-l, i, and i+l, of a multi-bit binary addend signal a. and a multi-bit binary augend signal b to be added (where i could be a positive or negative number designating the denominational order of the particular bit in question) . Each stage comprises two full adders (CSA) . Each full adder is preferably imple¬ mented as described in FIGS. 1 and 2. Addend signal a. comprises two component binary signals, a sum signal and a carry signal. The sum of these component signals represents one of the binary numbers to be added. Augend signal b also comprises two component binary signals, a sum signal and a carry signal. The sum of these component signals represents the other binary number to be added.

Although the binary signals are represented as single ended, in practice they are preferably bipolar in nature so as to utilize the logic circuitry described

above. Each stage of the adder has first and second full adders, preferably of the type disclosed in FIG. 2, each CSA having complementary input terminals X, Y and C^, and output terminals C ₀ and S. At each stage, one bit of the complementary sum component signal a is applied to complementary input terminals X, the corresponding bit of the complementary carry component signal a is applied to complementary input terminals Y, and the corresponding bit of the complementary sum component signal b is applied to complementary input terminals C _j _ of the first CSA. The complementary signals at output terminals C ₀ of one bit of the first CSA are applied to complementary input terminals Cj_ of the second CSA of the stage for the next higher order bit. The complementary signals at output terminals S of the first CSA are applied to complementary input terminals X of the second CSA of the same stage. The corresponding bit of the complementary carry component signal b is applied to complementary input terminals Y of the second CSA. (Alternatively, the carry component signal b could be applied to input terminal C _j _ of the first CSA and the sum component signal b could be applied to input terminal Y of the second CSA.) Each bit of the resultant signal R generated by the adder comprises a sum component and a carry component as represented in FIG. 3 by Rj_-1, Rj_, Rj_+1. For each stage, the complementary signals at output terminal S comprise the sum component of the resultant signal for the bit, e.g., the "i" bit, and the complementary signals generated at output terminal C ₀ of the next lower order stage, e.g., the "i-l" bit, comprise the carry component of the resultant signal for that bit.

In summary, the entire addition process is completed without propagation of carries with an execution time that is independent of the word length of the addend and augend signals. Specifically, the execution time is the

sum of the delay times through the first CSA and the second CSA.

The invention contemplates a digital computing system that employs a plurality of the adders described in connection with FIG. 3 connected in tandem to perform arithmetic operations such as addition, multiplication, and accumulation. FIG. 4 illustrates a typical digital computing system utilizing the adders of FIG. 3. A multi-bit binary addend signal a. and a multi-bit binary augend signal b are applied- to an arithmetic unit, i.e., an adder 80 constructed as shown in FIG. 3. Similarly, a multi-bit binary addend signal d and a multi-bit binary augend signal e are applied to another arithmetic unit, i.e., an adder 82 constructed as shown in FIG. 3. Typically, since binary signals a, b, d, and .e are input signals to the computing system, they would not have a carry component, although they could. Assuming that they do not have a carry component, one input of the first CSA and one input of the second CSA of each stage, e«g- input terminal Y in FIG. 3, would not be used, or a binary carry component signal representing the number zero would be applied to these input terminals. Adder 80 generates a multi-bit binary component sum signal r-_ and a multi-bit binary component carry signal c-_ . The sum of signals r^ and c^ represents the sum of the numbers represented by signals a. and b. Similarly, adder 82 generates a multi-bit binary component sum signal r2 and a multi-bit binary component carry signal c ₂ .

Signals r _l7 c^, r ₂ , and c ₂ are applied to another arithmetic unit, i.e., a multiplier 84. Multiplier 84 can be constructed from a number of adders of the type shown in FIG. 3 as building blocks using conventional techniques for digital multiplication such as, for example, the so-called Booth algorithm. ^' Generally, multiplier 84 would have a number of adders of the type shown in FIG.

3 equal to the bits in the numbers being multiplied. Signal r^ serves as a sum component multiplicand signal, signal c^ serves as a carrier component multiplicand signal, signal r serves as a sum component multiplier signal, and signal c ₂ serves as a carrier component multiplier signal. Multiplier 84 generates a multi-bit binary component product signal r3 and a multi-bit binary component carry signal C ₃ . Each adder used in multiplier 84 separately adds a binary sum component signal and a binary carry component signal to another binary signal (with or without components) to produce a separate binary component sum signal and a binary component carry signal. As a result, the time required to complete the multipli¬ cation process s dramatically reduced vis-a-vis a con- ventional multiplier because the carry bits do not prop¬ agate from stage to stage. The more bits in the binary signals being processed, the greater the reduction in processing time vis-a-vis a conventional arithmetic unit because the processing time is independent of the number of bits in the binary signals being ^' processe ^' d.

Binary signal r ₃ is applied to an adder 85 as a component addend signal, binary signal C3 is applied to adder 85 as a component addend signal, and a binary augend signal f_ is applied to adder 85 to be additively combined with the component addend signals. Depending upon the source of signal .f, i.e., whether or not it was the subject of previous processing in the computing system, it could also comprise two component signals. In either case, adder 85, configured as shown in FIG. 3, generates a multi-bit binary component sum signal r ₄ and a multi-bit binary component carry signal C . The sum of the numbers represented by signals r ₄ and c ₄ represents the sum of the numbers represented by signals r3, 03, and f.

Binary signals r ₄ and c ₄ are applied to an accumulator

86. Accumulator 86, which comprises the adder shown in FIG. 3, an output register for storing sum and carry components of such an adder, and feedback connections to the inputs of the adder, performs the conventional accumulating function of repeatedly adding the sum stored in the output register to the input signal applied to the accumulator. Signal r ₄ serves as one component of the signal to be accumulated, and signal c ₄ serves as the other component of the signal to be accumulated. Accumulator 86 generates a multi-bit binary component signal r ₅ , which is the accumulation of the component sum signals generated by the adder, and a multi-bit binary component signal C5, which is the accumulation of the component carry signals generated by the adder. As a result, the time required to complete the accumulation process is dramatically reduced vis-a-vis a conventional accumulator because the carry, bits do not propagate from stage to stage. The more bits in the binary signals being processed, the greater the reduction in processing time vis-a-vis a conventional arithmetic unit because the processing time is independent of the number of bits in the binary signals being processed.

Signals τc _$ and C5 are applied to a conventional adder 88, which forms a single, multi-bit binary sum signal Rf. In summary, although the illustrated computing system employs numerous adders in performing the arithmetic operations of addition, multiplication and accumulation, only one adder, namely, adder 88, incurs the delay occasioned by propagation of carries through its stages. Thus, the processing time is dramatically reduced.

The use of complementary signals and sum and carry components increases the wire interconnections fourfold relative to a conventional carry-propagating adder implemented with single-ended logic circuitry. In order

to minimize the complexity of the wire interconnections between stages, the stages of each adder are laid out on an integrated circuit chip in physically contiguous relationship to each other in bit order, and the same order stages of tandemly connected adders are laid out on the integrated circuit chip in physically aligned relationship with each other. This is illustrated in FIG. 5 where an adder 90 is connected in tandem with an adjacent adder 92. Adder 90 has individual adder stages for processing a "0" bit, -a "1" bit, a "2" bit, . . . and an "n" bit. Adder 92 has individual adder stages for processing a "0" bit, a "1" bit, a "2" bit, . . . and an "n" bit. Each adder stage is preferably constructed as shown in FIGS. 2 and 3. Other circuit components of an arithmetic unit, such as registers, multiplier related logic circuitry, and accumulator feedback connections, have been omitted for simplicity. These would also be individualized on the integrated circuit chip for each stage and, thus, be laid out in physically contiguous relationship to each other in bit order and in physically aligned relationship with the other circuit components of the same bit. In other words, the circuit components are laid out on the chip in a grid of rows and columns- - the rows comprising the same order bits of different circuit components and the columns comprising different order bits of the same circuit components.

As illustrated, the stages of adder 90 and the stages of adder 92, respectively, are physically contiguous to each other in bit order, e.g., the "0" bit stage is contiguous to the "1" bit stage, and the "1" bit stage is contiguous to the "2" bit stage, etc. Similarly, the same order stages of adjacent, tandemly-connected adders are physically aligned with each other, e.g., the "0" bit stage of adder 90 is physically aligned with the "0" bit stage of adder 92, and the "1" bit stage of adder 90 is

physically aligned with the "1" bit stage of adder 92, etc. As used herein, the term "aligned" means either directly aligned or aligned with a constant offset.

FIG. 5 also illustrates the number of signals pro- cessed by the stages of adders 90 and 92. Adder 90 operates upon a multi-bit binary addend signal a and a multi-bit binary augend signal b. Each bit of signal a and signal b comprises two bipolar, i.e., complementary, component (e.g., sum and carry) signals. Each bit of signal a and each bit of 'signal b are applied to the corresponding bit stage of adder 90, e.g., signals a- _| _ and bi are applied to the "1" bit stage of adder 90. Thus, as indicated at a^_ and as indicated at b]_, four augend signals are applied to the "1" bit stage of adder 90. In other words, a total of eight signals are applied to each stage of adder 90. Adder 90 generates a multi- bit binary signal x that represents the sum of the addend and augend signals. Each stage generates four bipolar, i.e., complementary, component (e.g. sum and carry) signals. The component sum signal from each stage is used with the component carry signal from the next lower bit stage to form the input to the same order stage of the adjacent tandemly-connected adder, e.g. , the sum component from the "1" bit stage and the carry component from the ^ιι o ^» bit stage of adder 90 together form the four signals designated x± , and the ^" sum component from the "2" bit stage and the carry component from the "1" bit stage of adder 92 together form the four signals designated y ₂ . A multi-bit binary addend signal- d is processed with signal x in adder 92. As illustrated, each bit of signal d comprises two bipolar, i.e., complementary, component (e.g., sum and carry) signals. Each bit of signal d and each bit of signal x are applied to the corresponding bit stage of adder 92, e.g., signals d ₂ and x ₂ are applied to the "2" bit stage of adder 92.