Bipolar Integrated Technology, Inc.

Bruce E Miller

Robert E Owen

Edward Bass

Steven T Reyburn

Felice E Bailey

Dean M Ford

An integrated circuit multiplier-accumulator architecture includes an M-bit wide register for inputting an X operand and an N-bit wide input register for inputting a Y operand to a multiplier. The multiplier can selectably multiply or concatenate the operands to produce a binary product in the form of a first array of M+N parallel bits. A binary adder adds the binary product to a second array of M+N+P+1 parallel bits and outputs the sum as a Z result in the form of a third array of M+N+P+1 parallel bits. The Z result is stored in a selected one of two accumulators. A feedback path is provided to output selected accumulator contents to the adder as the second binary array of M+N+P+1 bits. Output ports are provided for outputting a selected portion of the accumulator contents. Preferably, the output ports can output the entire M+N+P bits in parallel, as well as any selected portion thereof. Overflow logic can be provided which determines from the (M+ n+P+1)th bit whether an overflow has occurred in the M+N+P bit result. A format adjust circuit is provided between the accumulators and the output ports for shifting the entire output accumulator contents a predetermined number of bits within a range of zero to at least P bits, and preferably P+1 bits, in the direction of the most significant bit.

Fixed-point multiplier-accumulator architecture

A high-speed digital multiplier architecture is implemented in a bipolar very large scale integrated circuit technology. Operand input and product output latches are independently enabled by inverted clock signals. The multiplier can be operated in unclocked, separately clocked and single clock or master-slave modes of operation. The multiplier can be operated to concatenate, rather than multiply, the operands and thereby load the operands directly from the inputs to the output. A selectable format adjust performs a one bit left shift on the product. A low order zero bit is inserted in the shifted product, an overflow flag is set in case the product is -1.0.times.-1.0=1.0, and rounding is correct for both adjusted and unadjusted products. A zero flag is provided which is correct for both rounded and unrounded output products. A negative flag provides an unambiguous indicator of product sign in signed and mixed mode or format adjusted operation.

Growing community of inventors

Aloha, OR, United States of America

Bruce E Miller