Floating-Point Multiplier Rounding Circuit for Shorter Critical Paths

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The rounding calculation in prior art always adds to the critical path in some fashion, making it inefficient for floating point multiplication, which results in a temporary mantissa with more bits of precision than can be stored, requiring specific rounding modes to be defined for IEEE floating point specifications.

Innovation Solution

The solution simultaneously calculates the mantissa values for no rounding, rounding up with no additional left shift, and rounding up with an additional left shift, reducing the critical timing path by combining instances where the 2K term may be added, and using RTL code to generate the mantissa values efficiently.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If rounding calculation is performed sequentially after multiplication, then correct IEEE floating point rounding is achieved, but the critical path delay increases

Engineering Contradiction:
Improvecorrect roundingVSAvoidcritical path delay
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent calculates the three possible mantissa values (no rounding, round up with no shift, round up with shift) in parallel during the multiplication operation itself, rather than sequentially after. This preliminary calculation of all possible rounding outcomes eliminates the need for sequential rounding operations, reducing critical path delay while ensuring correct IEEE floating point rounding through selective use of pre-calculated values based on the rounding mode and exponent comparison

Inventive Principle:
Principle #10Preliminary action

2Adaptability or versatility

If multiple rounding mode calculations are performed separately, then all IEEE rounding modes are supported, but the circuit complexity and timing path increase

Engineering Contradiction:
Improverounding mode supportVSAvoidcircuit complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent merges the calculation of three different rounding outcomes into a single unified circuit structure. Instead of implementing separate rounding circuits for each IEEE rounding mode (round to nearest, round toward zero, round toward positive infinity, round toward negative infinity), the invention calculates all possible mantissa values simultaneously using shared hardware resources, including common adders and multiplexers that select the appropriate result based on the rounding mode and exponent comparison, thereby reducing overall circuit complexity

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentUS8832166B2Floating point multiplier circuit with optimized rounding calculation
Publication Date: 2014.09.09 TEXAS INSTRUMENTS INC
  • US8832166B2 patent drawing
  • US8832166B2 patent drawing
  • US8832166B2 patent drawing

AI summary

An optimized floating point multiplier rounding circuit that minimizes the increase of the critical timing path of the calculation. The values of the temporary mantissa required to make the rounding decision are calculated simultaneously by the circuit shown in the invention.