Abstract:Modern day computing increasingly relies on specialization to satiate growing performance and efficiency requirements. A core challenge in designing such specialized hardware architectures is how to perform mapping space search, i.e., search for an optimal mapping from algorithm to hardware. Prior work shows that choosing an inefficient mapping can lead to multiplicative-factor efficiency overheads. Additionally, the search space is not only large but also non-convex and non-smooth, precluding advanced search techniques. As a result, previous works are forced to implement mapping space search using expert choices or sub-optimal search heuristics. This work proposes Mind Mappings, a novel gradient-based search method for algorithm-accelerator mapping space search. The key idea is to derive a smooth, differentiable approximation to the otherwise non-smooth, non-convex search space. With a smooth, differentiable approximation, we can leverage efficient gradient-based search algorithms to find high-quality mappings. We extensively compare Mind Mappings to black-box optimization schemes used in prior work. When tasked to find mappings for two important workloads (CNN and MTTKRP), the proposed search finds mappings that achieve an average $1.40\times$, $1.76\times$, and $1.29\times$ (when run for a fixed number of steps) and $3.16\times$, $4.19\times$, and $2.90\times$ (when run for a fixed amount of time) better energy-delay product (EDP) relative to Simulated Annealing, Genetic Algorithms and Reinforcement Learning, respectively. Meanwhile, Mind Mappings returns mappings with only $5.32\times$ higher EDP than a possibly unachievable theoretical lower-bound, indicating proximity to the global optima.

EasyMap: Improving Technology Mapping Via Exploration-Enhanced Heuristics and Adaptive Sequencing

MapTune: Advancing ASIC Technology Mapping via Reinforcement Learning Guided Library Tuning

MacroMap: A Technology Mapping Algorithm for Heterogeneous FPGAs with Effective Area Estimation

Improvements to Technology Mapping for LUT-Based FPGAs

Heuristics for Area Minimization in LUT-Based FPGA Technology Mapping

FPGA Technology Mapping with Adaptive Gate Decomposition.

Dual-output LUT merging during FPGA technology mapping

Parallelized Technology Mapping to General PLBs by Adaptive Circuit Partitioning

FPGA Technology Mapping Using Sketch-Guided Program Synthesis

MEC: An Open-source Fine-grained Mapping Equivalence Checking Tool for FPGA

FlowMap: an optimal technology mapping algorithm for delay optimization in lookup-table based FPGA designs

FMAP: a Technology Mapping Algorithm for FPGA with MUX-LUT Mixed Architecture

On area/depth trade-off in LUT-based FPGA technology mapping

Mind Mappings: Enabling Efficient Algorithm-Accelerator Mapping Space Search

An Improved Algorithm for Performance Optimal Technology Mapping with Retiming in LUT-Based FPGA Desig

Power Optimization Through Edge Reduction In Lut-Based Fpga Technology Mapping

Physically Aware Synthesis Revisited: Guiding Technology Mapping with Primitive Logic Gate Placement

LUT-based FPGA Technology Mapping: A Simulated Annealing Approach

PBMap: A Path Balancing Technology Mapping Algorithm for Single Flux Quantum Logic Circuits

PLBMAP:A High Performance and Universal Programmable Logic Block Mapping Algorithm Based on FPGA

An FPGA Robust Technology Mapping Method