Skip to content
STIMSMITH

FPGA Acceleration

Technique WIKI v1 · 8/6/2026

FPGA acceleration is a hardware acceleration technique that uses Field-Programmable Gate Arrays to speed up computationally intensive workloads. The provided evidence focuses specifically on its use in accelerating hardware verification flows (functional verification of Chisel-built processor models and FPGA-accelerated fuzzing frameworks such as HWFuzz for RISC-V processor verification), where it shortens verification cycles by moving stimulus generation, execution, and coverage collection onto programmable logic.

FPGA Acceleration

FPGA acceleration refers to the use of Field-Programmable Gate Arrays (FPGAs) to accelerate compute- or verification-intensive tasks that would otherwise run more slowly on general-purpose processors. Within the scope of the supplied evidence, the technique is discussed chiefly as a means to accelerate hardware design verification workflows.

Use in Processor Model Verification (Chisel / ARM)

Wu, Wang, and Chen describe a functional verification methodology for an ARM V4 instruction-set compatible processor model, ARMChisel, built with the Chisel hardware construction language. Their methodology comprises:

  1. A random instruction generator covering the full ARM V4 ISA, used to produce test stimuli.
  2. Four verification stages—Chisel-level primary verification, rapid coverage-driven verification, directed test verification, and complex application verification—designed around the characteristics of the Chisel language.
  3. Testbenches constructed in both the Chisel and Verilog environments that collect coverage and rapidly localize errors.

As a final step, the authors apply an FPGA acceleration method to accelerate the verification of large-scale application programs and shorten the verification cycle. [1]

Use in FPGA-Accelerated Fuzzing (HWFuzz)

HWFuzz is an FPGA-accelerated fuzzing framework for RISC-V processor verification that addresses the performance bottleneck caused by running the hardware fuzzer in software. Its architecture exploits the Zynq UltraScale+ device, which integrates Programmable Logic (PL) with a Processing System (PS) on the same chip:

  • Programmable Logic (PL): Hosts the hardware fuzzer IP and the RISC-V processor design-under-test (DUT).
  • Processing System (PS): Runs the corresponding software reference model of the DUT.
  • ENCORE-style differential checking dynamically compares DUT execution against the reference model.

The hardware fuzzer IP operates in two coverage-directed modes:

  • Random mode: Selects instructions purely at random.
  • Mutation mode: Adjusts operands and context of stimuli previously generated in random mode that have the highest probability of expanding DUT coverage, storing generated stimuli and corresponding DUT coverage data in a hardware-side corpus.

HWFuzz can also automatically instrument fine-grained synthesizable coverpoints into the DUT during compilation, allowing coverage information to be collected directly from the FPGA at runtime and fed back to the fuzzer. Generated instructions and data are stored in a designated DDR region that serves as the testing environment for the DUT. [2]

Common Pattern Across Evidence

Both pieces of evidence share a common pattern: FPGA acceleration is applied to verification of processor designs, where it:

  • Speeds up stimulus generation and execution relative to software-only simulation.
  • Enables coverage collection from synthesized hardware running in real time.
  • Shortens the overall verification cycle for large-scale workloads.

Related Entities

  • Research on functional verification method processor model built by Chisel (Paper): Uses FPGA Acceleration as the final step to accelerate large-scale program verification and shorten the verification cycle. [1]

References from Public Context

The provided public context also contains arXiv works on FPGA acceleration in adjacent domains—photoacoustic image reconstruction and binary CNN accelerators—that illustrate the broader applicability of the technique but are not directly cited as sources for the claims above.

LINKED ENTITIES

1 links
Research on functional verification method processor model built by Chisel USES

CITATIONS

4 sources
4 citations
[1] FPGA acceleration is used to accelerate verification of large-scale application programs for a Chisel-built ARM V4 processor model, shortening the verification cycle. Research on functional verification method processor model built by Chisel
[2] HWFuzz is an FPGA-accelerated fuzzing framework for RISC-V processor verification that uses the Zynq UltraScale+ PL/PS architecture to host the fuzzer IP and DUT on PL while the software reference model runs on PS. HWFuzz: An FPGA-Accelerated Fuzzing Framework for Efficient RISC-V Processor Verification
[3] The HWFuzz hardware fuzzer IP operates in random and mutation modes to perform coverage-directed stimulus generation, with stimuli and corresponding DUT coverage stored in a hardware-side corpus. HWFuzz: An FPGA-Accelerated Fuzzing Framework for Efficient RISC-V Processor Verification
[4] HWFuzz can automatically instrument fine-grained synthesizable coverpoints into the DUT during compilation so coverage information is collected directly from the FPGA at runtime. HWFuzz: An FPGA-Accelerated Fuzzing Framework for Efficient RISC-V Processor Verification