Leveraging Highly Approximated Multipliers in DNN Inference
This work addresses power efficiency in DNN accelerators for hardware deployment, but it is incremental as it builds on existing approximate multiplier methods.
The paper tackled the problem of error in DNN inference caused by using approximate multipliers for power savings, and the result was a control variate technique that achieved 45% power reduction with less than 1% accuracy loss and improved accuracy by 1.9x compared to approximate designs without the technique.
In this work, we present a control variate approximation technique that enables the exploitation of highly approximate multipliers in Deep Neural Network (DNN) accelerators. Our approach does not require retraining and significantly decreases the induced error due to approximate multiplications, improving the overall inference accuracy. As a result, our approach enables satisfying tight accuracy loss constraints while boosting the power savings. Our experimental evaluation, across six different DNNs and several approximate multipliers, demonstrates the versatility of our approach and shows that compared to the accurate design, our control variate approximation achieves the same performance, 45% power reduction, and less than 1% average accuracy loss. Compared to the corresponding approximate designs without using our technique, our approach improves the accuracy by 1.9x on average.