PECAN: A Product-Quantized Content Addressable Memory Network
This work addresses hardware efficiency for deep neural networks, particularly in in-memory computing, though it appears incremental as it builds on existing quantization and memory techniques.
The paper tackles the problem of hardware-efficient deep neural network deployment by proposing PECAN, a novel architecture that uses product quantization for filtering and linear transforms, enabling implementation via content addressable memory with simple table lookups. Experiments confirm its feasibility, offering a multiplier-free solution with implications for in-memory computing.
A novel deep neural network (DNN) architecture is proposed wherein the filtering and linear transform are realized solely with product quantization (PQ). This results in a natural implementation via content addressable memory (CAM), which transcends regular DNN layer operations and requires only simple table lookup. Two schemes are developed for the end-to-end PQ prototype training, namely, through angle- and distance-based similarities, which differ in their multiplicative and additive natures with different complexity-accuracy tradeoffs. Even more, the distance-based scheme constitutes a truly multiplier-free DNN solution. Experiments confirm the feasibility of such Product-Quantized Content Addressable Memory Network (PECAN), which has strong implication on hardware-efficient deployments especially for in-memory computing.