Agentic Electronic Design Automation: A Handoff Perspective

Abstract: Electronic design automation (EDA) is inherently multi-stage and handoff-heavy. Design artifacts, flow scripts, and engineering decisions cross tool, session, and organizational boundaries before final implementation, signoff, or release. Each transfer carries explicit and implicit requirements that may not be fully captured by stage-local checks. LLM-based agents now invoke EDA tools directly, embed retrieved knowledge in executable scripts, and hand off state across sessions and stages. Once their outputs condition downstream engineering decisions, the transferred object must satisfy a handoff contract and meet the assumptions of its next consumer. This survey introduces handoff validity as its organizing principle. A handoff is valid when the transferred object satisfies the consumer's acceptance conditions and carries sufficient context, evidence, and provenance for downstream use. We review 82 systems and classify them into three boundary classes. Stage-Bound systems establish validity within a single EDA stage or bounded verification task. Flow-Bound systems preserve coherent workflow state across tools, invocations, and sessions. Organization-Bound systems maintain source grounding, provenance, scope, and admissibility across knowledge and authority boundaries. For each class, we analyze handoff contracts, handoff objects, coordination mechanisms, and open questions. These analyses motivate a five-layer EDA agent communication protocol (EACP), covering the agent discovery, agent message, tool invocation, workflow orchestration, and security and IP protocols. We aim to provide a common vocabulary and research agenda for trustworthy agentic EDA.

• The paper introduces handoff validity as a core concept to ensure transferred objects meet downstream acceptance across EDA stages and organizations.
• The paper categorizes agentic EDA systems into Stage-Bound, Flow-Bound, and Organization-Bound classes, detailing distinct coordination and validation mechanisms.
• The paper proposes a five-layer EDA Agent Communication Protocol aimed at enhancing interoperability, auditability, and IP protection in agentic workflows.

Agentic Electronic Design Automation from the Handoff Perspective

Introduction

"Agentic Electronic Design Automation: A Handoff Perspective" (2606.19795) systematically analyzes the structural challenges in Electronic Design Automation (EDA) arising from the proliferation of LLM-based agents operating across heterogeneous toolchains, stages, and organizational contexts. Rather than focusing solely on localized automation, the paper frames EDA workflows as fundamentally handoff-heavy, punctuated by transfers of artifacts, scripts, constraints, and decisions that cross operational and authority boundaries. The core organizing principle introduced is handoff validity: the transferred object must satisfy the downstream consumer’s acceptance conditions and carry sufficient context, evidence, and provenance for it to be reliably reused. The taxonomy developed centers not on application or modality, but on the boundary of validity preservation—divided into Stage-Bound, Flow-Bound, and Organization-Bound classes.

Boundary-Centered Taxonomy and Handoff Validity

The survey critically distinguishes three classes of agentic EDA systems:

• Stage-Bound: Validity is enforced within a single EDA stage via local checkers (functional, specification, performance, or physical verification). Artifact transfers rarely encode all cross-stage dependencies, exposing latent incompatibility and context loss risks.
• Flow-Bound: Validity must persist across tool invocations, pipeline stages, sessions, and optimization rounds. Downstream failures often originate upstream, and maintaining coherent workflow state, consistent constraints, and QoR records is non-trivial.
• Organization-Bound: Validity depends on source grounding, provenance, scope matching, and admissibility across knowledge and authority boundaries. The challenge lies in traceability and contextual compatibility for answers, scripts, or procedural actions.

Each boundary class is analyzed along four orthogonal dimensions: handoff contracts (acceptance and evidence), handoff objects (artifacts, state, commands), coordination mechanisms (generation, orchestration, retrieval), and open questions (context representation, oracle composition, recovery, provenance).

Detailed Analysis of Agentic EDA System Classes

Stage-Bound Systems

Stage-Bound mechanisms primarily operate under local acceptance contracts, closing repair or optimization loops using executable artifacts and checkers. The surveyed literature documents robust pipelines for RTL repair [9, 24], analog sizing [25, 80], and testbench or assertion generation [27, 77]. Strong numerical outcomes are reported for syntactic and simulation closure, but the handoff rarely encodes full downstream assumptions.

Hybrid systems integrate multiple oracles (e.g., compilers plus physical verification) and employ nuanced coordination (generate-repair, search-optimize, supporting). However, the boundary between mechanistic closure and actual downstream usability remains porous. Open technical challenges include representation and prioritization of heterogeneous feedback, composition of local oracles, and sufficiency criteria for stage-local closure.

Flow-Bound Systems

Flow-Bound systems orchestrate inter-tool and inter-session continuity, focusing on preserving executable scripts, configuration validity, and downstream measurements (QoR records). Complex multi-agent frameworks for ASIC integration [20], flow optimization [4], and tool orchestration [3, 82] require stateful recovery and coordinated transitions. Systems such as SoCDev [84] and Autonomous Evolution [88] demonstrate resilient checkpointing, rollback, and QoR-driven optimization, with strong empirical evidence of cross-session reliability and reproducibility.

Hybrid mechanisms combine orchestrate-transform, recover-resume, and annotate-optimize patterns. Unresolved issues include minimal valid context bundles for cross-stage handoff, attribution of downstream failures, and aggregation of distributed evidence for closure. The literature illustrates that execution correctness does not guarantee continuity at the flow level, and comprehensive validation protocols are absent.

Organization-Bound Systems

Organization-Bound systems address retrieval, grounding, and admissibility challenges for knowledge and command artifacts crossing team and authority boundaries. Documented frameworks (Ask-EDA [5], ChatLS [13], ChipXplore [30]) employ hybrid retrieval and graph-based reasoning to ensure answers and scripts are attributable and scoped correctly. Strong results emerge in semantic retrieval and source traceability, but organizational acceptance—comprising scope matching, provenance, and authorization—is not fully enforced.

Coordination mechanisms (retrieve-answer, retrieve-execute) are supplemented by multi-modal and graph-based evidence linking. Challenges persist in encoding and propagating authority scope, defining granularity for provenance chains, and validating grounded commands against policy, syntax, and current design context.

EDA Agent Communication Protocol (EACP) and Implications

The lack of standardized interoperability and contract verification motivates the proposed five-layer EDA Agent Communication Protocol:

• L1 (Agent Discovery): Registry and capability manifest encoding process, tool, artifact, and contract compatibility.
• L2 (Agent Message): Structured handoff message preserving intent, action, context, evidence, and correlation.
• L3 (Tool Invocation): Typed operation descriptors mapping standard actions to backend-specific tools, with pre-/postconditions and error semantics.
• L4 (Workflow Orchestration): Directed workflow graphs with explicit dependency, transition, checkpoint, and recovery semantics.
• L5 (Security & IP): Authorization, disclosure, protection, retention, and audit trail policies enabling accountable handoff across organizational boundaries.

The architectural proposal reflects industry developments (e.g., Synopsys AgentEngineer, Cadence ChipStack AI Super Agent) but addresses gaps in cross-vendor semantic interchange and IP protection. EACP aims to provide reference types (ContextRef, ArtifactRef, EvidenceRef, PrincipalRef, PolicyRef, TraceRef) sharing state across all protocol layers, critically necessary for portable and auditable design artifacts.

This layered protocol, if realized, would fundamentally change agent-based EDA workflows, enabling modular interoperability, rigorous validation, and accountable IP management across heterogeneous environments and organizational structures.

Conclusion

The survey delivers a comprehensive boundary-centered framework for analyzing agentic EDA systems, emphasizing that robust local automation does not guarantee validity and portability across design, flow, and organizational boundaries. The taxonomy and proposed EACP clarify the structural requirements for trustworthy and interoperable agent-based design automation. The implications are significant for both research and industrial practice: future agentic EDA solutions must encode and validate cross-boundary contracts, preserve full context and provenance, and support system-level auditability and IP protection. Long-term, this foundation is critical for deploying agentic workflows that are not only technically effective but also accountable and admissible in multi-party collaborative settings.