An Ultra-Efficient Approximate Multiplier With Error Compensation for Error-Resilient Applications

Approximate computing is a promising approach for enhancing hardware efficiency in error-tolerant applications like neural networks and image processing by sacrificing some accuracy. This brief introduces an exceptionally efficient approximate multiplier equipped with error compensation capabilities. The proposed multiplier involves a constant compensation term for the least significant half of the product, while the other half is precisely calculated, striking a balance between hardware efficiency and accuracy. Additionally, a low-complexity yet effective error compensation module (ECM) is presented, significantly enhancing overall accuracy. Through simulation in VIVADO and comparison with exact and existing approximate designs, the results demonstrate that the proposed multiplier achieves high accuracy comparable to exact multipliers in neural networks. Consequently, it stands as a viable alternative for exact multipliers in real-world error-resilient applications.

Index Terms—Error compensation, approximate multiplier, neural network, hardware-accuracy trade-off.