Skip to content
STIMSMITH

GenHuzz

Tool
First seen 6/10/2026
Last seen 9/1/2026
Evidence 46 chunks

NEIGHBORHOOD

43 nodes · 87 edges
graph · GenHuzz · depth=1

RELATIONSHIPS

50 connections
Rocket Chip evaluates → 100% 5e
GenHuzz is evaluated on RocketChip, identifying new bugs and vulnerabilities.
CVA6 evaluates → 100% 5e
GenHuzz is evaluated on CVA6, identifying new bugs and vulnerabilities including PMP bypass.
TheHuzz compares with → 100% 4e
GenHuzz is benchmarked against TheHuzz on coverage metrics.
Hardware-Guided Reinforcement Learning implements → 100% 4e
GenHuzz implements the Hardware-Guided Reinforcement Learning framework to dynamically optimize its fuzzing policy based on hardware feedback.
BOOM evaluates → 100% 4e
GenHuzz is evaluated on Boom, identifying new bugs and vulnerabilities.
ChatFuzz ← compares with 100% 4e
GenHuzz is benchmarked against ChatFuzz on coverage metrics.
Golden Reference Model uses → 100% 4e
GenHuzz uses the Golden Reference Model as ground truth for correct execution to detect bugs.
fuzzing policy uses → 100% 4e
GenHuzz optimizes a fuzzing policy through HGRL to generate effective test cases.
Reward Function uses → 100% 4e
GenHuzz uses a reward function combining validity, hardware coverage, and bonus rewards.
Reinforcement Learning uses → 100% 4e
GenHuzz uses reinforcement learning to fine-tune its fuzzing policy based on hardware coverage feedback.
DiFuzzRTL ← compares with 100% 4e
GenHuzz is benchmarked against DifuzzRTL on coverage metrics.
White-Box Fuzzing implements → 100% 4e
GenHuzz is described as a white-box hardware fuzzing framework.
Cascade ← compares with 100% 3e
GenHuzz is benchmarked against Cascade on coverage metrics across three RISC-V cores.
hardware fuzzing implements → 100% 3e
GenHuzz is a hardware fuzzing framework targeting RISC-V hardware cores.
GenHuzz uses a reset module to balance exploration and exploitation when the fuzzing policy converges.
RISC-V assembly instruction generation uses → 100% 3e
GenHuzz generates RISC-V assembly instructions as test cases for hardware fuzzing.
coverage-based fuzzing implements → 100% 3e
GenHuzz is described as a coverage-based white-box fuzzing method that maximizes exploration of hardware states.
GenHuzz: An Efficient Generative Hardware Fuzzer ← introduces 100% 3e
The GenHuzz paper introduces the GenHuzz tool as an efficient generative hardware fuzzer.
Load-Reserved/Store-Conditional instructions mentions → 95% 2e
GenHuzz detects vulnerabilities related to Load-Reserved/Store-Conditional instructions on Boom and Rocket.
RISC-V assembly generation implements → 100% 2e
GenHuzz generates RISC-V assembly instructions as test cases using its language model-based fuzzer.
language model-based fuzzer implements → 100% 2e
GenHuzz employs a language model-based fuzzer to generate RISC-V assembly instructions.
Proximal Policy Optimization uses → 100% 2e
GenHuzz uses PPO to bridge the fuzzer-DUT interaction in the HGRL framework.
spike uses → 100% 2e
GenHuzz uses Spike as the RISC-V ISA reference model (Golden Reference Model) for differential testing.
Synopsys VCS uses → 100% 2e
GenHuzz uses Synopsys VCS to generate hardware simulation traces and compute hardware coverage metrics.
hardware coverage uses → 100% 2e
GenHuzz uses hardware coverage as feedback to guide test case generation.
device under test uses → 100% 2e
GenHuzz interacts with the Device Under Test to obtain hardware coverage feedback.
Instruction Semantics uses → 100% 2e
GenHuzz comprehends instruction semantics to generate effective test cases.
Generative Pre-trained Transformer uses → 100% 2e
GenHuzz leverages a GPT-based (Generative Pre-trained Transformer) language model as its fuzzer.
Physical Memory Protection evaluates → 95% 2e
GenHuzz detected vulnerabilities in the Physical Memory Protection implementation of RISC-V cores.
Load-Reserved Store-Conditional evaluates → 95% 2e
GenHuzz detected vulnerabilities related to Load-Reserved/Store-Conditional instructions in Boom and Rocket cores.
Testcase Generation uses → 100% 2e
GenHuzz uses test case generation to create RISC-V assembly instruction sequences for hardware fuzzing.
Finite State Machine Coverage uses → 100% 2e
GenHuzz uses FSM coverage as one of the hardware coverage metrics for evaluation.
hardware coverage feedback uses → 100% 2e
GenHuzz uses hardware coverage feedback from the DUT to refine its fuzzing policy.
differential testing uses → 100% 2e
GenHuzz uses differential testing to detect bugs by comparing RTL and GRM execution traces.
GPT-based language model fuzzer uses → 100% 2e
GenHuzz leverages a GPT-based language model as both a fuzzer and scorer.
inter-instruction semantics uses → 95% 2e
GenHuzz understands inter-instruction semantics to strategically combine multiple instructions.
RTL Simulation uses → 95% 2e
GenHuzz uses RTL simulation as part of differential testing to detect hardware bugs.
mismatch detection uses → 100% 2e
GenHuzz uses automated mismatch detection to identify discrepancies between DUT and GRM execution traces.
instruction tokenization and encoding uses → 95% 2e
GenHuzz tokenizes and encodes instructions into numerical representations for model training.
Execution Trace Comparison uses → 100% 2e
GenHuzz compares execution traces from DUT and Spike to detect bugs.
Spike RISC-V ISA Simulator uses → 100% 2e
GenHuzz uses Spike as the RISC-V ISA reference model for differential testing.
Physical Memory Protection mentions → 95% 2e
GenHuzz detects vulnerabilities related to Physical Memory Protection configurations.
The paper mentions GenHuzz as a related ML-based fuzzing work.
HiFuzz compares with → 90% 1e
HiFuzz is structurally compared with GenHuzz as a related generative fuzzer.
Lichao Wu authored by → 100% 1e
Lichao Wu is listed as an author of GenHuzz.
Mohamadreza Rostami authored by → 100% 1e
Mohamadreza Rostami is listed as an author of GenHuzz.
Huimin Li authored by → 100% 1e
Huimin Li is listed as an author of GenHuzz.
Jeyavijayan Rajendran authored by → 100% 1e
Jeyavijayan Rajendran is listed as an author of GenHuzz.
Ahmad-Reza Sadeghi authored by → 100% 1e
Ahmad-Reza Sadeghi is listed as an author of GenHuzz.
Byte-Pair Encoding mentions → 90% 1e
GenHuzz mentions that GPT-2's Byte-Pair Encoding is unsuitable for RISC-V assembly and replaces it with word-level encoding.