Deep neural networks (DNNs) have achieved great breakthroughs in the past decades. However, DNNs require massive multiply-accumulate (MAC) operations and their execution on digital hardware, e.g., GPUs, causes enormous energy consumption. This poses critical risks to performance and energy sustainability in Artificial Intelligence (AI) computing systems. To enhance computational and energy efficiency, state-of-the-art research proposes to use logic gates to construct neural networks directly. It was demonstrated that such a neural network can achieve higher computational and energy efficiency than traditional neural networks with massive multiply-accumulate operations. The concept of logic gate neural network is shown in the following figure.

In this thesis, look-up-tables (LUTs) will be used to construct neural networks from scratch to enhance the logic gate neural networks. Such LUT-based neural networks will be validated on FPGAs to demonstrate the advantages of such neural networks in inference accuracy, computational and energy efficiency.