AritPIM: High-Throughput In-Memory Arithmetic
share
Digital processing-in-memory (PIM) architectures are rapidly emerging to overcome the memory-wall bottleneck by integrating logic within memory elements. Such architectures provide vast computational power within the memory itself in the form of parallel bitwise logic operations. We develop novel algorithmic techniques for PIM that, combined with new perspectives on computer arithmetic, extend this bitwise parallelism to the four fundamental arithmetic operations (addition, subtraction, multiplication, and division), for both fixed-point and floating-point numbers, and using both bit-serial and bit-parallel approaches. We propose a state-of-the-art suite of arithmetic algorithms, demonstrating the first algorithm in the literature for a majority of cases - including cases previously considered impossible for digital PIM, such as floating-point addition. Through a case study on memristive PIM, we compare the proposed algorithms to an NVIDIA RTX 3070 GPU and demonstrate massive throughput and energy improvements. Overall, this paper provides a fundamental foundation for high-throughput arithmetic in PIM, thereby also exposing PIM algorithmic research to the wider computer-science community as all fundamental arithmetic operations can now be abstractly used, on both fixed and floating-point numbers.
READ FULL TEXT