Skip to content

Basit-Balogun10/isqed-2026-dv-challenge

Repository files navigation

Agentic AI Design Verification Challenge 2026

ISQED 2026 | March 31 -- April 7, 2026

A week-long competition where teams build AI-assisted verification environments for real-complexity RTL designs derived from the OpenTitan ecosystem. Participants develop testbenches, drive coverage closure, debug failures, and build autonomous verification agents.

This is a Design Verification competition -- not a security CTF. The battlefield is testbenches, coverage models, constrained-random stimulus, assertions, and the relentless grind toward coverage closure.


Quick Start

# Clone the repo
git clone https://github.com/<org>/isqed-2026-dv-challenge.git
cd isqed-2026-dv-challenge

# Verify your toolchain (Path A -- open-source)
iverilog -g2012 -o /dev/null duts/common/dv_common_pkg.sv duts/nexus_uart/nexus_uart.sv

# Install cocotb
pip install cocotb cocotb-bus cocotb-coverage

# Run a skeleton testbench
cd skeleton_envs/nexus_uart
make SIM=icarus

All DUTs use flat TL-UL signals (no packed structs) for maximum simulator compatibility. See the Instructions & Documentation for the full TL-UL signal list.

No pre-built environments are provided. Participants set up their own tools. Teams without local EDA tools may use EDA Playground for free access to commercial simulators.


Competition Structure

Phase Title Window Duration
Phase 1 Verification Environment Construction Mar 31 -- Apr 2 72 hrs
Phase 2 Coverage Closure Campaign Apr 2 -- Apr 4 48 hrs
Phase 3 Debug & Root-Cause Analysis Apr 4 -- Apr 5 36 hrs
Phase 4 Agentic Verification Pipeline Apr 5 -- Apr 7 36 hrs

Phases overlap slightly -- you can start Phase N+1 while finishing Phase N.


DUT Portfolio

Seven designs of increasing complexity, all using a common TileLink-UL bus interface:

DUT Tier Description Scoring Weight
nexus_uart Starter Full-duplex UART with FIFOs, parity, loopback 1.0x
bastion_gpio Starter 32-pin GPIO with edge detection, masked writes 1.0x
warden_timer Intermediate 64-bit timer + watchdog with bark/bite 1.5x
citadel_spi Intermediate SPI host, 4 modes, segment-based commands 1.5x
aegis_aes Advanced AES-128 ECB/CBC encrypt/decrypt engine 2.0x
sentinel_hmac Advanced HMAC-SHA256 with streaming message FIFO 2.0x
rampart_i2c Expert Dual-role I2C host/target, multi-master, clock stretching 2.5x

Tasks

Phase 1 -- Verification Environment Construction

Task Name Max Points Description
1.1 Interface Agent Generation 900 Build bus and protocol agents (driver, monitor, scoreboard)
1.2 Verification Plan to Test Suite 1200 Generate tests from YAML verification plans
1.3 Register Verification Suite 200 Auto-generate CSR regression tests from Hjson maps
1.4 Assertion Library 700 Write assertions that catch bugs without false positives

Phase 2 -- Coverage Closure Campaign

Task Name Max Points Description
2.1 Coverage Gap Analysis 200 Analyze and prioritize uncovered regions
2.2 Stimulus Engineering 2000 Write tests to close coverage gaps
2.3 Coverage-Directed Test Generation 300 Build an adaptive AI-in-the-loop CDG system

Phase 3 -- Debug & Root-Cause Analysis

Task Name Max Points Description
3.1 The Log Whisperer 300 Triage simulation failures from logs
3.2 The Trace Detective 400 Waveform-guided debug and root-cause tracing
3.3 The Regression Medic 300 Bucket failures, validate patches

Phase 4 -- Agentic Verification Pipeline

Task Name Max Points Description
4.1 The AutoVerifier 1000 Build an autonomous end-to-end verification agent
4.2 The Regression Agent 600 CI/CD verification agent for RTL commit streams
4.3 The Spec Interpreter 500 Generate verification plans from specs alone

Total possible: ~8,600+ points


Repository Contents

├── duts/                  RTL source (SystemVerilog) for all 7 DUTs
│   └── common/            Shared TL-UL bus interface package
├── specs/                 Specification documents per DUT (markdown)
├── csr_maps/              Register maps in Hjson format
├── vplans/                Verification plans per DUT (YAML)
├── skeleton_envs/         Starter cocotb testbenches with TL-UL agent
├── phase3_materials/      Debug & root-cause analysis inputs (Phase 3)
│   ├── task_3_1_logs/     10 simulation failure logs
│   ├── task_3_2_traces/   5 failure traces with signal data
│   └── task_3_3_regression/  Regression suite (200 tests, 35 failing, 5 patches)
└── platform_content/      Competition docs and task descriptions
    ├── competition_rules.md
    ├── instructions_and_documentation.md
    ├── submission_requirements.md
    ├── evaluation/        Scoring rubrics
    └── tasks/             Extended task descriptions (linked above)

Two Verification Paths

Path A (Open-Source) Path B (Commercial)
Simulator Verilator + Icarus Verilog VCS / Questa / Xcelium
Testbench cocotb + pyuvm (Python) SystemVerilog UVM
Coverage cocotb-coverage Native SV covergroups
Assertions SVA via bind files Full SVA support
Constrained Random Manual / Python random Native solver (rand, constraint)
Starter code Provided Build your own
License needed No Yes

Both paths are fully supported and scored identically. See the Instructions for setup details.


Submission

Submissions are made through the competition website (linked from the ISQED 2026 event page). Multiple submissions per task are allowed -- only your highest score counts.

Each submission must include a methodology.md documenting your AI usage (20% of score).

Key documents:


Scoring

Pillar Weight Measured By
Correctness 30% Bug detection rate, false positive rate (automated)
Coverage 30% Line, branch, toggle, FSM, functional coverage (automated)
Methodology 20% Code quality, architecture, verification best practices (judges)
AI Integration 20% Innovation and effectiveness of AI usage (judges)

Full scoring rubric: evaluation_rubrics.md


RTL Disclaimer

The RTL designs provided in this repository are modified versions of open-source IP blocks, created exclusively for this competition. They are not production-quality, not security-reviewed, and must not be used outside of this competition context. Signal names and internal architectures have been modified from their original sources.


License

Competition materials are provided for use in the ISQED 2026 Agentic AI Design Verification Challenge only. See individual files for applicable licenses.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors