arXiv Computer Architecture Papers - Curated Collection

arXiv's computer architecture listing is where processor verification research appears months before it reaches a conference, and where the open RISC-V ecosystem publishes work that never goes anywhere else. This is a short, deliberately selective list: papers I have read that say something a verification engineer can use, each with a summary of what it does and why it matters, not a restatement of the abstract.

What ties the list together. Most of it is about verifying RISC-V cores faster or more thoroughly than a software testbench allows: FPGA-accelerated co-verification against a reference model, learned stimulus generation replacing random mutation, instruction-set-simulator lockstep for superscalar pipelines, and security assertions that can be carried from one core to the next. The rest is architectural context, on pipeline depth, virtualization and the RISC-V security landscape, chosen because you cannot write a good test plan for a feature you do not understand.

Editorial note: summaries are my own reading. Where a number is quoted, it is the authors' measurement; the assessment of what it means is mine.

9

Papers

5

Years

18

Topics

How to use: Browse by year, or use Ctrl+F to search for topics like "RISC-V", "Verification", "Pipeline", "Cache", etc.

2025(5 papers)

Title	Summary	Link
FERIVer: FPGA-assisted Emulated Framework for RTL Verification of RISC-V Processors RISC-VVerificationFPGARTL	Instruction-set simulators are fast but approximate; RTL simulation is exact but slow, and processor verification needs both. FERIVer puts the RISC-V core under test into the programmable fabric of a Zynq SoC and runs a reference model on the embedded CPU, cross-checking the two at 5 million instructions per second. The authors report roughly 150x the throughput of Xilinx XSim and 35x that of Verilator, using under 7 percent of the FPGA. The interesting part for a DV engineer is the co-verification plumbing: how the two sides stay in lockstep and how mismatches are captured with enough context to debug.	View Paper
Lyra: Hardware-Accelerated RISC-V Verification with Generative Model-Based Fuzzing RISC-VVerificationFuzzingAI/ML	Hardware fuzzing borrowed mutation-based test generation from software, but random byte mutations of an instruction stream produce shallow programs that converge slowly on coverage. Lyra keeps the hardware-accelerated part, running the RISC-V DUT and its reference model concurrently on an FPGA SoC for differential checking and coverage collection, and replaces the mutator with an ISA-aware generative model trained to produce semantically valid, coverage-directed instruction sequences. It is the clearest example so far of learned stimulus generation feeding a conventional differential testbench, and the coverage-convergence curves are the numbers to look at.	View Paper
Translating Common Security Assertions Across Processor Designs: A RISC-V Case Study RISC-VSecurityAssertionsVerification	Assertion-based security verification works, but writing the assertions for each new core is expensive. This paper proposes translating an existing library of security assertions from one processor to another with a different microarchitecture, focused on RISC-V. The method maps signals and properties across designs so that checks written for one core can be regenerated for the next with little manual effort. If you maintain SVA libraries across a family of cores, the mapping approach is directly reusable even outside the security context.	View Paper
Wildcat: Educational RISC-V Microprocessors RISC-VPipelineEducation	Every architecture course teaches the five-stage pipeline, a choice rooted in 1980s technology. Wildcat asks whether shorter pipelines are better for teaching and for small embedded cores, and implements RISC-V variants with three, four and five stages on FPGA and in the open SkyWater 130 nm ASIC flow. Measured by maximum clock frequency and resource cost, the longer pipeline does not always win. A useful reference if you are building a teaching core or a tiny controller and want data rather than folklore behind the pipeline depth.	View Paper
Flexing RISC-V Instruction Subset Processors to Extreme Edge RISC-VEdge ComputingVerification	Flexible integrated circuits for extreme-edge devices cannot afford a full RISC-V core, so this paper generates instruction-subset processors automatically from an application's instruction profile. Each instruction is implemented as a discrete, pre-verified hardware block, and the generator composes only the blocks the application needs. Verification is therefore front-loaded into the block library rather than repeated per processor. The methodology is a concrete instance of correct-by-construction composition and is worth reading for the verification argument alone.	View Paper

2024(1 paper)

Title	Summary	Link
SupeRFIVe: Functional ISS-Driven Verification of Superscalar RISC-V Processors RISC-VVerificationSuperscalarISS	SupeRFIVe verifies a dual-issue superscalar RISC-V core by connecting the RTL testbench to the Spike instruction set simulator over a socket and comparing architectural state as benchmark programs execute. The contribution is the methodology for keeping an ISS and a superscalar pipeline in step, where instructions retire out of the simple one-per-cycle order that lockstep comparison usually assumes. If you have used or built an ISS-driven checker for an in-order core, this is the next step up in difficulty.	View Paper

2023(1 paper)

Title	Summary	Link
CVA6 RISC-V Virtualization: Architecture, Microarchitecture, and Design Space Exploration RISC-VVirtualizationCVA6	A complete account of adding hardware virtualization, the RISC-V hypervisor extension, to the open-source CVA6 core: the architectural additions, the microarchitecture, including a G-stage TLB and an L2 TLB to keep two-level address translation affordable, and a design-space exploration with post-layout results in 22 nm FD-SOI. Verification engineers will find the description of two-stage translation and its corner cases a good primer before writing tests for any core with a hypervisor mode.	View Paper

2021(1 paper)

Title	Summary	Link
A Survey on RISC-V Security: Hardware and Architecture RISC-VSecuritySurvey	RISC-V is displacing Arm in embedded and IoT designs, and with it comes the need for a root of trust and protection against tampering and leakage. This survey organizes the RISC-V security landscape: hardware vulnerabilities, ISA and architectural security features, trusted execution proposals, and verification approaches including formal methods such as UPEC. It is the best single starting point for understanding what a security verification plan for a RISC-V SoC has to cover.	View Paper

2018(1 paper)

Title	Summary	Link
Trends in Processor Architecture A. González PipelineCacheMicroarchitectureSurvey	A concise overview of how processor architecture evolved and why: the driving forces behind deeper pipelines, cache hierarchies, branch prediction, out-of-order execution and the move to multicore, followed by a discussion of promising directions. Short enough to read in one sitting and a good refresher for verification engineers who want the architectural reasons behind the microarchitectural features they are asked to verify.	View Paper

---

All papers are open-access via arXiv cs.AR.

Last updated: 2026-09-20

Know a great paper? Leave a comment below!

Related Posts

• DVCon Papers - Curated Collection (Papers)
• AI in Design Verification: What DVCon 2025–2026 Tells Us (AI, DVCon, Papers)