Multi-Stage Floating-Point Accumulator Pipeline

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The existing floating-point accumulation process in digital processors is limited by the time required for operand alignment, summation, and post-alignment, which restricts the clock speed and overall processing speed, especially in applications that extensively perform floating-point operations.

Innovation Solution

A multi-stage floating-point accumulator is introduced, partitioning the accumulation process into multiple stages with each stage having a fraction of the total delay, utilizing latches and multiplexers to align and sum operands, and a 3-operand adder to perform the accumulation, allowing the use of accumulated results from earlier clock cycles.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If floating-point accumulation is performed in a single stage, then the circuit structure is simple, but the total delay is large and clock speed is limited

Engineering Contradiction:
Improveclock speedVSAvoidcircuit structure
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The floating-point accumulation process is divided into multiple stages (first stage, second stage, third stage, etc.), where each stage performs a subset of the alignment and addition operations. This segmentation reduces the delay within each stage, enabling higher clock speeds while maintaining the complete accumulation functionality across all stages.

Inventive Principle:
Principle #1Segmentation

2Loss of time

If the accumulation process includes complete alignment and post-alignment steps, then calculation accuracy is maintained, but the time required for each operation increases

Engineering Contradiction:
Improveoperation timeVSAvoidcalculation accuracy
Core Design Contradiction:
Loss of timeVSMeasurement precision

Solution Approach 1:

Operand alignment is performed in advance during the first stage before the addition operations occur in subsequent stages. This preliminary alignment ensures that the mantissas are properly positioned for accurate addition, while the alignment results are preserved through latches between stages, maintaining calculation accuracy without requiring repeated alignment operations.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP2057535B1Multi-stage floating-point accumulator
Publication Date: 2010.11.10 QUALCOMM INC
  • EP2057535B1 patent drawingFigure 1
  • EP2057535B1 patent drawingFigure 2
  • EP2057535B1 patent drawingFigure 3

AI summary

A multi-stage floating-point accumulator includes at least two stages and is capable of operating at higher speed. In one design, the floating-point accumulator includes first and second stages. The first stage includes three operand alignment units, two multiplexers, and three latches. The three operand alignment units operate on a current floating-point value, a prior floating-point value, and a prior accumulated value. A first multiplexer provides zero or the prior floating-point value to the second operand alignment unit. A second multiplexer provides zero or the prior accumulated value to the third operand alignment unit. The three latches couple to the three operand alignment units. The second stage includes a 3-operand adder to sum the operands generated by the three operand alignment units, a latch, and a post alignment unit.