Pith. sign in

REVIEW 1 major objections 42 references

ZK-Tracer: A High-Performance Heterogeneous Accelerator for Zero-Knowledge VM Trace Generation

T0 review · 1 major / 0 minor · reviewed 2026-06-29 · grok-4.3

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A heterogeneous ASIC called ZK-Tracer speeds zkVM trace generation by up to 1829 times over a high-performance CPU.

desk verdict The abstract claims 1829x trace generation speedup and 963x end-to-end but supplies zero details on baselines, workloads, or measurement method. read the letter →

arxiv 2605.25493 v2 pith:4Q4VPTYB submitted 2026-05-25 cs.AR

Jieran Cui , Zhengkai Wen , Haowen Fang , Yinan Zhu , Jia Xiong , Cheng Ni , Mingchi Zhang , Nan Guan
show 1 more author
Xi Wang
This is my paper · ORCID
classification cs.AR
keywords zero-knowledgeproofszkVMhardwareacceleratortracegenerationASICheterogeneousarchitecturefrontendaccelerationpermutationtraces
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper identifies the frontend execution and trace generation phase as the emerging bottleneck in zero-knowledge virtual machines. It introduces ZK-Tracer, a hardware accelerator with a main trace unit and multiple parallel permutation trace units that receives tasks through a lightweight instruction set extension from the host. ASIC measurements show the design delivers the stated speedups in trace generation and, when paired with existing backend accelerators, produces a 963x improvement in overall ZKP system throughput. A sympathetic reader would care because removing this frontend limit could make zkVM-based proofs viable for larger workloads.

What carries the argument

Heterogeneous accelerator built from one Main Trace Unit plus parallel Permutation Trace Units, controlled through a lightweight instruction-set extension for host offloading.

What would settle it

A benchmark run on current zkVM workloads that shows trace generation occupying less than 10 percent of total ZKP runtime, or an ASIC measurement that fails to exceed 100x speedup in trace generation.

Watch

Extended reading notes

Core claim

ZK-Tracer is the first accelerator architecture built specifically for the zkVM frontend. Its heterogeneous layout consists of a Main Trace Unit handling core execution traces and parallel Permutation Trace Units handling the permutation traces required by the proof system. The unit exposes a fine-grained interface via a lightweight instruction set extension so that software can offload trace tasks efficiently. Fabricated ASIC results establish the 1829x trace-generation speedup over multi-core CPU and the 963x end-to-end system gain when the accelerator is combined with existing proving hardware.

Load-bearing premise

The frontend execution and trace generation phase has become the dominant performance bottleneck in zkVM systems.

Editorial extensions

If this is right

  • Trace generation ceases to limit zkVM throughput once the accelerator is present.
  • End-to-end ZKP pipelines achieve roughly three orders of magnitude higher performance when the frontend accelerator is added to existing backend hardware.
  • zkVMs become practical for workloads whose size was previously ruled out by frontend cost.
  • Software can be restructured around the new offload interface to keep the accelerator fed.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Similar trace-generation hardware could be adapted to other proof systems that rely on execution traces.
  • The reported speedups assume the host CPU remains the source of program execution; a fully integrated design might change the bottleneck again.
  • The 963x end-to-end figure depends on the relative sizes of frontend and backend workloads in a given application.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 0 minor

Summary. The manuscript proposes ZK-Tracer, the first hardware accelerator architecture for the zkVM frontend trace generation phase. It introduces a heterogeneous design with a Main Trace Unit and parallel Permutation Trace Units, exposed to host software via a lightweight instruction set extension for task offloading. ASIC implementation results are claimed to deliver up to 1829× speedup in trace generation versus a high-performance multi-core CPU and 963× end-to-end improvement for the full ZKP system when paired with existing backend accelerators.

Significance. If the reported speedups can be substantiated with complete benchmark descriptions, workloads, baselines, and measurement details, the work would be significant for identifying and addressing the frontend as an emerging bottleneck in zkVMs. The heterogeneous architecture and integration approach, if validated, could enable practical system-level gains in zero-knowledge proof generation.

major comments (1)
  1. [Abstract] Abstract: the central performance claims (1829× trace-generation speedup and 963× end-to-end improvement) are stated without any description of benchmarks, workloads, measurement methodology, error bars, comparison baselines, ASIC process node, or host offload/data-movement costs. This absence renders the primary empirical results unevaluable and is load-bearing for the paper's contribution.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comment on the abstract. We agree that additional context is needed to make the central claims evaluable and will revise the manuscript to address this.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central performance claims (1829× trace-generation speedup and 963× end-to-end improvement) are stated without any description of benchmarks, workloads, measurement methodology, error bars, comparison baselines, ASIC process node, or host offload/data-movement costs. This absence renders the primary empirical results unevaluable and is load-bearing for the paper's contribution.

    Authors: We agree the abstract is too terse on evaluation details. In the revised manuscript we will expand the abstract to briefly specify: the workloads (representative zkVM programs including Fibonacci, sorting, and SHA256 circuits), the CPU baseline (high-performance 64-core Xeon-class processor), the ASIC process node, and that reported end-to-end figures include host offload and data-movement overheads. Full methodology, run counts, and any variance will continue to appear in the evaluation section. This change directly addresses the evaluability concern while preserving abstract length. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: empirical hardware benchmarks only

full rationale

The paper reports measured ASIC performance results (1829x trace-generation speedup and 963x end-to-end) against external CPU and backend baselines. No equations, derivations, fitted parameters, predictions, or first-principles claims appear in the abstract or described architecture. The load-bearing steps are direct empirical measurements, not reductions to self-defined quantities or self-citations, so the result is self-contained against external benchmarks.

Assumptions & free parameters 0 free parameters · 0 assumptions · 0 invented entities

Abstract contains no mathematical model, free parameters, axioms, or postulated entities; the contribution is an architectural description and performance claim only.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ZK-Tracer: A High-Performance Heterogeneous Accelerator for Zero-Knowledge VM Trace Generation." pith.science (2026). https://pith.science/paper/4Q4VPTYB

@misc{pith2026260525493,
  author       = {Pith},
  title        = {Pith review of: ZK-Tracer: A High-Performance Heterogeneous Accelerator for Zero-Knowledge VM Trace Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4Q4VPTYB}},
  note         = {Machine review of arXiv:2605.25493}
}
read the original abstract

Zero-knowledge virtual machines (zkVMs) are a key technology for driving the large-scale adoption of zero-knowledge proofs (ZKP), but their performance bottlenecks severely limit their practicality. While current hardware acceleration research has exclusively focused on backend proving, we identify that the frontend execution and trace generation phase is rapidly emerging as the new system bottleneck. To address this challenge, we propose ZK-Tracer, the first hardware accelerator architecture specifically designed for the zkVM frontend. ZK-Tracer features a novel heterogeneous design comprising a Main Trace Unit and parallel Permutation Trace Units. It exposes a fine-grained interface to the host software through a lightweight instruction set extension, enabling efficient task offloading. Our ASIC implementation results demonstrate that ZK-Tracer achieves up to 1829x speedup in trace generation over a high-performance multi-core CPU. When integrated with existing backend proving accelerators, it delivers a remarkable 963x end-to-end performance improvement for the entire ZKP system.

Figures

Figures reproduced from arXiv: 2605.25493 by the authors.

Figure 1
Figure 1. zkVM Workflow in [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. zkVM Trace Generation Flow Permutation Trace Units (PTUs) for parallel processing of the Permutation Trace. These units work efficiently through a custom Instruction Set Architecture (ISA). • We introduce a suite of algorithm-hardware co-design inno￾vations that exploit the mathematical properties and struc￾tural characteristics of ZKP traces. Key techniques include: (1) zero-overhead trace capture via non-intrusive… view at source ↗
Figure 3
Figure 3. Profiling before (left) and after backend acceleration (right) [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (5 more)
Figure 5
Figure 5. Figure 5: Architecture of ZK-Tracer microarchitectural extensions for non-intrusive trace capture, and Instruction Set Architecture (ISA) customizations for fine-grained, software-defined control. The Main Trace Unit (MTU) is responsi￾ble for accurately executing the zkVM RISC-V…
Figure 6
Figure 6. Figure 6: Main Trace Unit activity. By inserting these instructions to bracket a target code seg￾ment, developers or compilers can precisely demarcate the region of interest for proving. This fine-grained control mechanism of￾fers significant flexibility, drastically reducing un…
Figure 7
Figure 7. Figure 7: MMAC Systolic Array [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]
Figure 8
Figure 8. Figure 8: Batch Modular Inverse Unit property of zkVM traces: they are “narrow and long” ,where the number of columns, 𝑘, is typically small (from a few to hundreds), while the number of rows, 𝑁, can be in the millions or more. We discard the inefficient method of recomputing we…
Figure 9
Figure 9. Figure 9: Parallelism Analysis FibonacciIs_Prime Gorth16_VerifyRSA BLS12-381 BN254 SHA256Tendermint 0 250 500 750 1000 1250 1500 1750 2000 Speedup MTU Avg: 315× PTU Avg: 1514× MTU Speedup PTU Speedup [PITH_FULL_IMAGE:figures/full_fig_p006_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references

  1. [1]

    Anees Ahmed, Nojan Sheybani, Davi Moreno, Nges Brian Njungle, Tengkai Gong, Michel Kinsy, and Farinaz Koushanfar. 2024. Amaze: Accelerated mimc hardware architecture for zero-knowledge applications on the edge. InProceedings of the 43rd IEEE/ACM International Conference on Computer-Aided Design. 1–9

    2024

  2. [2]

    Bing-Jyue Chen, Suppakit Waiwitlikhit, Ion Stoica, and Daniel Kang. 2024. Zkml: An optimizing system for ml inference in zero-knowledge proofs. InProceedings of the Nineteenth European Conference on Computer Systems. 560–574

    2024

  3. [3]

    Alhad Daftardar, Jianqiao Mo, Joey Ah-kiow, Benedikt Bünz, Ramesh Karri, Siddharth Garg, and Brandon Reagen. 2025. Need for zkSpeed: Accelerating HyperPlonk for Zero-Knowledge Proofs. InProceedings of the 52nd Annual Inter- national Symposium on Computer Architecture. 1986–2001

    2025

  4. [4]

    Alhad Daftardar, Brandon Reagen, and Siddharth Garg. 2024. Szkp: A scalable accelerator architecture for zero-knowledge proofs. InProceedings of the 2024 International Conference on Parallel Architectures and Compilation Techniques. 271–283

    2024

  5. [5]

    George Danezis, Cedric Fournet, Markulf Kohlweiss, and Bryan Parno. 2013. Pinocchio coin: building zerocoin from a succinct pairing-based proof system. In Proceedings of the First ACM workshop on Language support for privacy-enhancing technologies. 27–30

    2013

  6. [6]

    Antoine Delignat-Lavaud, Cédric Fournet, Markulf Kohlweiss, and Bryan Parno

  7. [7]

    509 certificates into elegant anonymous credentials with the magic of verifiable computation

    Cinderella: Turning shabby X. 509 certificates into elegant anonymous credentials with the magic of verifiable computation. In2016 IEEE Symposium on Security and Privacy (SP). IEEE, 235–254

  8. [9]

    Boyuan Feng, Zheng Wang, Yuke Wang, Shu Yang, and Yufei Ding. 2024. Zeno: A type-based optimization framework for zero knowledge neural network infer- ence. InProceedings of the 29th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 1. 450–464

    2024

Show all 42 references
  1. [10]

    gem5. 2024. gem5. https://github.com/gem5/gem5. Version 24.1 Accessed: 2025-10-10

    2024

  2. [11]

    Shafi Goldwasser, Silvio Micali, and Chales Rackoff. 2019. The knowledge com- plexity of interactive proof-systems. InProviding sound foundations for cryptog- raphy: On the work of shafi goldwasser and silvio micali. 203–225

    2019

  3. [12]

    Florian Hirner, Ahmet Can Mert, and Sujoy Sinha Roy. 2024. Proteus: A pipelined ntt architecture generator.IEEE Transactions on Very Large Scale Integration (VLSI) Systems32, 7 (2024), 1228–1238

    2024

  4. [13]

    Zhuoran Ji, Zhiyuan Zhang, Jiming Xu, and Lei Ju. 2024. Accelerating multi- scalar multiplication for efficient zero knowledge proofs with multi-gpu systems. InProceedings of the 29th ACM International Conference on Architectural Support for Programming Languages and Operating...

    2024

  5. [14]

    Zhuoran Ji, Jianyu Zhao, Peimin Gao, Xiangkai Yin, and Lei Ju. 2025. Accelerat- ing Number Theoretic Transform with Multi-GPU Systems for Efficient Zero Knowledge Proof. InProceedings of the 30th ACM International Conference on Architectural Support for Programming Languages a...

    2025

  6. [15]

    Ahmed Kosba, Andrew Miller, Elaine Shi, Zikai Wen, and Charalampos Papa- manthou. 2016. Hawk: The blockchain model of cryptography and privacy- preserving smart contracts. In2016 IEEE symposium on security and privacy (SP). IEEE, 839–858

    2016

  7. [16]

    Shang Li, Zhiyuan Yang, Dhiraj Reddy, Ankur Srivastava, and Bruce Jacob. 2020. DRAMsim3: A cycle-accurate, thermal-capable DRAM simulator.IEEE Computer Architecture Letters19, 2 (2020), 106–109

    2020

  8. [17]

    Changxu Liu, Hao Zhou, Lan Yang, Yifei Feng, Zheng Wu, Zhuoyuan Yang, Yinlong Li, Shiyong Wu, and Fan Yang. 2025. AcclMT: A Highly Resource- Efficient and Flexible Poseidon Hash-Based Merkle Tree Architecture. In2025 62nd ACM/IEEE Design Automation Conference (DAC). IEEE, 1–7

    2025

  9. [18]

    Changxu Liu, Hao Zhou, Lan Yang, Zheng Wu, Patrick Dai, Yinlong Li, Shiyong Wu, and Fan Yang. 2024. Myosotis: An Efficiently Pipelined and Parameterized Multi-Scalar Multiplication Architecture via Data Sharing.IEEE Transactions on Computer-Aided Design of Integrated Circuits ...

    2024

  10. [19]

    Changxu Liu, Hao Zhou, Lan Yang, Jiamin Xu, Patrick Dai, and Fan Yang. 2024. Gypsophila: A scalable and bandwidth-optimized multi-scalar multiplication architecture. InProceedings of the 61st ACM/IEEE Design Automation Conference. 1–6

    2024

  11. [20]

    Tao Lu, Yuxun Chen, Zonghui Wang, Xiaohang Wang, Wenzhi Chen, and Ji- aheng Zhang. 2025. BatchZK: A Fully Pipelined GPU-Accelerated System for Batch Generation of Zero-Knowledge Proofs. InProceedings of the 30th ACM International Conference on Architectural Support for Program...

    2025

  12. [21]

    Tao Lu, Chengkun Wei, Ruijing Yu, Chaochao Chen, Wenjing Fang, Lei Wang, Zeke Wang, and Wenzhi Chen. 2023. Cuzk: Accelerating zero-knowledge proof with a faster parallel multi-scalar multiplication algorithm on gpus.IACR Transac- tions on Cryptographic Hardware and Embedded Sy...

    2023

  13. [22]

    Weiliang Ma, Qian Xiong, Xuanhua Shi, Xiaosong Ma, Hai Jin, Haozhao Kuang, Mingyu Gao, Ye Zhang, Haichen Shen, and Weifang Hu. 2023. Gzkp: A gpu accelerated zero-knowledge proof system. InProceedings of the 28th ACM Inter- national Conference on Architectural Support for Progr...

    2023

  14. [23]

    Bjorn Oude Roelink, Mohammed El-Hajj, and Dipti Sarmah. 2024. Systematic review: Comparing zk-SNARK, zk-STARK, and bulletproof protocols for privacy- preserving authentication.Security and Privacy7, 5 (2024), e401

    2024

  15. [24]

    Pengcheng Qiu, Guiming Wu, Tingqiang Chu, Changzheng Wei, Runzhou Luo, Ying Yan, Wei Wang, and Hui Zhang. 2024. MSMAC: Accelerating Multi-Scalar Multiplication for Zero-Knowledge Proof. InProceedings of the 61st ACM/IEEE Design Automation Conference. 1–6

    2024

  16. [25]

    Andy Ray, Benjamin Devlin, Fu Yong Quah, and Rahul Yesantharao. 2024. Hard- caml msm: A high-performance split cpu-fpga multi-scalar multiplication engine. InProceedings of the 2024 ACM/SIGDA International Symposium on Field Pro- grammable Gate Arrays. 33–39

    2024

  17. [26]

    RISC Zero. 2021. risc0. https://github.com/risc0/risc0. Accessed: 2025-11-16

    2021

  18. [27]

    Nikola Samardzic, Simon Langowski, Srinivas Devadas, and Daniel Sanchez. 2024. Accelerating zero-knowledge proofs through hardware-algorithm co-design. In 2024 57th IEEE/ACM International Symposium on Microarchitecture (MICRO). IEEE, 366–379

    2024

  19. [28]

    Eli Ben Sasson, Alessandro Chiesa, Christina Garman, Matthew Green, Ian Miers, Eran Tromer, and Madars Virza. 2014. Zerocash: Decentralized anonymous payments from bitcoin. In2014 IEEE symposium on security and privacy. IEEE, 459–474

    2014

  20. [29]

    succinctlabs. 2024. sp1. https://github.com/succinctlabs/sp1/tree/v1.0.1. Ac- cessed: 2025-11-10

    2024

  21. [30]

    2024.SP1 Technical Whitepaper

    succinctlabs. 2024.SP1 Technical Whitepaper. Technical Whitepa- per. succinctlabs. https://drive.google.com/file/d/1aTCELr2b2Kc1NS- wZ0YYLKdw1Y2HcLTr/view Accessed: 2025-11-10

    2024

  22. [31]

    syntacore. 2017. scr1. https://github.com/syntacore/scr1. Accessed: 2025-11-10

    2017

  23. [32]

    Justin Thaler. 2025. The path to secure and efficient zkVMs: How to track progress. https://a16zcrypto.com/posts/article/secure-efficient-zkvms-progress/

    2025

  24. [33]

    Cheng Wang and Mingyu Gao. 2023. Sam: A scalable accelerator for number theoretic transform using multi-dimensional decomposition. In2023 IEEE/ACM International Conference on Computer Aided Design (ICCAD). IEEE, 1–9

    2023

  25. [34]

    Cheng Wang and Mingyu Gao. 2025. UniZK: Accelerating Zero-Knowledge Proof with Unified Hardware and Flexible Kernel Mapping. InProceedings of the 30th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 1. 1101–1117

    2025

  26. [35]

    Xi Wang, John D Leidel, Brody Williams, Alan Ehret, Miguel Mark, Michel A Kinsy, and Yong Chen. 2021. xbgas: A global address space extension on risc-v for high performance computing. In2021 IEEE International Parallel and Distributed Processing Symposium (IPDPS). IEEE, 454–463

    2021

  27. [36]

    Xi Wang, Brody Williams, John D Leidel, Alan Ehret, Michel Kinsy, and Yong Chen. 2020. Remote atomic extension (rae) for scalable high performance com- puting. In2020 57th ACM/IEEE Design Automation Conference (DAC). IEEE, 1–6

    2020

  28. [37]

    Tiancheng Xie, Jiaheng Zhang, Zerui Cheng, Fan Zhang, Yupeng Zhang, Yongzheng Jia, Dan Boneh, and Dawn Song. 2022. zkbridge: Trustless cross-chain bridges made practical. InProceedings of the 2022 ACM SIGSAC Conference on Computer and Communications Security. 3003–3017

    2022

  29. [38]

    Yongkui Yang, Zhenyan Lu, Jingwei Zeng, Xingguo Liu, Xuehai Qian, and Zhibin Yu. 2024. Falic: An FPGA-Based Multi-Scalar Multiplication Accelerator for Zero-Knowledge Proof.IEEE Trans. Comput.(2024)

    2024

  30. [39]

    Zhengbang Yang, Lutan Zhao, Peinan Li, Han Liu, Kai Li, Boyan Zhao, Dan Meng, and Rui Hou. 2025. LegoZK: A Dynamically Reconfigurable Accelerator for Zero- Knowledge Proof. In2025 IEEE International Symposium on High Performance Computer Architecture (HPCA). IEEE, 113–126

    2025

  31. [40]

    Jiaheng Zhang, Zhiyong Fang, Yupeng Zhang, and Dawn Song. 2020. Zero knowledge proofs for decision tree predictions and accuracy. InProceedings of the 2020 ACM SIGSAC Conference on Computer and Communications Security. 2039–2053

    2020

  32. [41]

    Ye Zhang, Shuo Wang, Xian Zhang, Jiangbin Dong, Xingzhong Mao, Fan Long, Cong Wang, Dong Zhou, Mingyu Gao, and Guangyu Sun. 2021. Pipezk: Accel- erating zero-knowledge proof with a pipelined architecture. In2021 ACM/IEEE 48th Annual International Symposium on Computer Architec...

    2021

  33. [42]

    Hao Zhou, Changxu Liu, Lan Yang, Li Shang, and Fan Yang. 2024. A Fully Pipelined Reconfigurable Montgomery Modular Multiplier Supporting Variable Bit-Widths.IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems43, 12 (2024), 4653–4665

    2024

  34. [43]

    Hao Zhou, Changxu Liu, Lan Yang, Li Shang, and Fan Yang. 2024. ReZK: A Highly Reconfigurable Accelerator for Zero-Knowledge Proof.IEEE Transactions on Circuits and Systems I: Regular Papers(2024)

    2024

Pith tools

Reviewed June 29, 2026 · model on record in the stance chip above.