Basic Operations And Structure Of An FPGA Accelerator For Parallel Bit Pattern Computation

Abstract

Parallel Bit Pattern computing (PBP) has been proposed as a way to dramatically reduce power consumption per computation by minimizing the total number of gate operations. In part, this reduction is accomplished by employing aggressive compiler optimization technology to gate-level representations of computations at runtime. Massive SIMD parallelism is used to obtain speedups while executing the optimized bit-serial code. However, the PBP model also can potentially exponentially reduce the number of active gates for each such operation by recognizing and operating on symbolically-compressed patterns of bits, rather than on each individual bit within a vector. This not only provides for efficient execution of traditional parallel code, but by using bit vectors to represent entangled superposition, enables quantum-like computation to be efficiently implemented using conventional circuitry. Building on lessons learned from various software and Verilog prototypes, this paper proposes a new set of basic operations and interface structure suitable for using inexpensive Xilinx Zynq-7000 boards to implement FGPA-hardware-accelerated PBP computation. Emphasis is on how these operations will implement quantum-like computation, as the first prototype system is currently still under development.