HEAM: High-Efficiency Approximate Multiplier Optimization for Deep Neural Networks

01/20/2022

∙

We propose an optimization method for the automatic design of approximate multipliers, which minimizes the average error according to the operand distributions. Our multiplier achieves up to 50.24 best reproduced approximate multiplier in DNNs, with 15.76 25.05 multiplier, our multiplier reduces the area, power consumption, and delay by 44.94 tested DNN accelerator modules with our multiplier obtain up to 18.70 area and 9.99

READ FULL TEXT

HEAM: High-Efficiency Approximate Multiplier Optimization for Deep Neural Networks

Low Error-Rate Approximate Multiplier Design for DNNs with Hardware-Driven Co-Optimization

Enabling Capsule Networks at the Edge through Approximate Softmax and Squash Operations

SemifreddoNets: Partially Frozen Neural Networks for Efficient Computer Vision Systems

PLAM: a Posit Logarithm-Approximate Multiplier for Power Efficient Posit-based DNNs

Deep Learning Training with Simulated Approximate Multipliers

scaleTRIM: Scalable TRuncation-Based Integer Approximate Multiplier with Linearization and Compensation

On the Approximation of Accuracy-configurable Sequential Multipliers via Segmented Carry Chains

HEAM: High-Efficiency Approximate Multiplier Optimization for Deep Neural Networks

Related Research

Low Error-Rate Approximate Multiplier Design for DNNs with Hardware-Driven Co-Optimization

Enabling Capsule Networks at the Edge through Approximate Softmax and Squash Operations

SemifreddoNets: Partially Frozen Neural Networks for Efficient Computer Vision Systems

PLAM: a Posit Logarithm-Approximate Multiplier for Power Efficient Posit-based DNNs

Deep Learning Training with Simulated Approximate Multipliers

scaleTRIM: Scalable TRuncation-Based Integer Approximate Multiplier with Linearization and Compensation

On the Approximation of Accuracy-configurable Sequential Multipliers via Segmented Carry Chains