Fast and Low-Power Quantized Fixed Posit High-Accuracy DNN Implementation

Sumit Walia, Bachu Varun Tej, Arpita Kabra, Joydeep Kumar Devnath, Joycee Mekie

2021IEEE Transactions on Very Large Scale Integration (VLSI) Systems15 citationsDOI

Abstract

This brief compares quantized float-point representation in posit and fixed-posit formats for a wide variety of pre-trained deep neural networks (DNNs). We observe that fixed-posit representation is far more suitable for DNNs as it results in a faster and low-power computation circuit. We show that accuracy remains within the range of 0.3% and 0.57% of top-1 accuracy for posit and fixed-posit quantization. We further show that the posit-based multiplier requires higher power-delay-product (PDP) and area, whereas fixed-posit reduces PDP and area consumption by 71% and 36%, respectively, compared to (Devnath <i>et al.</i>, 2020) for the same bit-width.

Topics & Concepts

Fixed pointQuantization (signal processing)Computer scienceMultiplier (economics)Fixed-point arithmeticComputationPower consumptionArtificial neural networkAlgorithmPower (physics)Floating pointArtificial intelligenceMathematicsEconomicsMacroeconomicsPhysicsQuantum mechanicsMathematical analysisAdvanced Neural Network ApplicationsModel Reduction and Neural NetworksAdversarial Robustness in Machine Learning