Tuple Tech
IC Design AI

Assertions that secure silicon

Avestra Studio

Transform specifications and RTL into sign-off-ready SystemVerilog Assertions using an 8-agent orchestrated pipeline — in minutes, not weeks.

Drop your spec (.pdf, .docx) and RTL (.sv, .v, .vh, .svh) — Avestra generates a complete SVA suite and testbench.

SystemVerilog Assertions<1% hallucination rateSign-off ready8-agent pipelineQuesta · VCS · XceliumLLM-agnostic

<1% hallucination rate

Avestra is trained on Ashok Mehta's 3 SVA textbooks, 22 US patents, and 35 years of CPU/ASIC/SoC domain expertise — domain intelligence no generic LLM can match. The result is deterministic, auditable output that holds up at sign-off.

<1%

hallucination rate
across SVA generation

From files to sign-off-ready assertions

Six stages, each powered by specialized agents — from spec parsing to final report.

1
RTL & Spec Analysis
Ingests RTL and specification files

Parses your RTL designs (.sv, .v, .vh, .svh) and specification documents (.pdf, .docx), extracting FSM states, temporal rules, clock domains, and cone-of-influence — building a structured model of your design intent.

2
Cross Check & Fix
Detects spec ↔ RTL inconsistencies

Compares the extracted spec model against RTL behavior. Flags undefined transitions, illegal state sequences, missing coverage targets, and timing violations — before any assertion is generated.

3
Assertion Generator
Synthesizes sign-off-ready SVA properties

Generates production-grade SystemVerilog Assertions — clocked assert, assume, and cover properties with vacuity guards, multi-clock support, and full source traceability back to spec and RTL.

4
Assertion Optimizer
Refines for coverage completeness

Reviews and tightens assertions for simulator compatibility and coverage completeness. Eliminates redundant properties, fills gaps, and validates output against known-good patterns from the proprietary SVA knowledge base.

5
Testbench Generator
Builds UVM-ready testbenches

Generates coverage-driven SystemVerilog testbenches with UVM-ready architecture, directed tests, and response checkers — targeting the specific coverage holes identified during analysis.

6
Final Session Report
Delivers a complete audit trail

Produces a full session report: all assertions, cover points, inconsistency findings, simulator targets (Questa/VCS/Xcelium), and a trace from every property back to its source in spec and RTL.

Eight agents, one pipeline, sign-off-grade output

Drop your spec and RTL files into Avestra Studio. The orchestrated pipeline runs end-to-end — catching inconsistencies, generating assertions, verifying coverage, and building testbenches — delivering a complete SVA suite ready for your simulator.

  • Accepts .pdf, .docx specs and .sv, .v, .vh, .svh RTL
  • Detects spec ↔ RTL inconsistencies before generation
  • Every assertion traced to its source in spec and RTL
  • UVM-ready testbench generated alongside the SVA suite
Avestra Studio · 8-Agent Pipeline · RunningAXI_slave.sv + spec.docx
1/8
Spec Analysis AgentFSM extracted · 14 temporal rules
2/8
RTL Analysis AgentCOI traced · 2 clock domains
3/8
RAG Context AgentDomain knowledge retrieved
4/8
Inconsistency Detection2 undefined transitions flagged
5/8
SVA Generation Agent18 properties synthesized
6/8
Coverage Analysis Agent6 cover points · vacuity verified
7/8
Cross-Check AgentAll assertions validated vs spec + RTL
8/8
Testbench Gen AgentUVM testbench + directed tests ready
Output Ready · Silicon-Grade
18 assertions · 6 cover points · SV testbench · inconsistency report
Questa·VCS·Xcelium

Everything you need for spec-to-SVA

Eight integrated capabilities across analysis, generation, optimization, and delivery — built on 35 years of SVA domain expertise.

8-agent orchestrated pipeline

Six user-visible steps powered by eight specialized agents under the hood — spec analysis, RTL analysis, RAG retrieval, inconsistency detection, SVA generation, coverage analysis, cross-check, and testbench generation.

Spec ↔ RTL inconsistency detection

Catches missing states, illegal transitions, and timing violations before the first simulation — before you spend hours debugging a broken testbench.

Production-grade SVA suite

Every generated assertion is clocked, vacuity-guarded, and traced to its source in spec and RTL. Multi-clocked and multi-reset designs are supported.

UVM-ready testbenches

Coverage-driven SV testbenches with UVM-ready architecture, directed tests targeting coverage holes, and response checkers — ready to drop into your existing simulation environment.

LLM-agnostic pipeline

Swap the underlying model without affecting output quality or structure. Works with frontier cloud models and self-hosted on-premise inference — keeping sensitive IP in your environment.

Simulator-independent output

Works with Questa (Siemens), VCS (Synopsys), and Xcelium (Cadence). No simulator vendor lock-in — same SVA suite across your full sign-off flow.

Broad input format support

Accepts spec documents as .pdf or .docx and RTL as .sv, .v, .vh, or .svh. Upload files directly in Avestra Studio — no preprocessing or format conversion required.

Proprietary SVA knowledge base

RAG trained on Ashok Mehta's 3 SVA textbooks and 22 US patents — 35 years of CPU/ASIC/SoC domain expertise encoded into every output. Deterministic, auditable, and sign-off-ready.

Ready to try Avestra Studio?

Early access is available now. Request access at avestra.net or book a live demo with the Tuple team to see the full pipeline on your design.