ETCH PROCESS WINDOW FACTORY AUTOMATION PORTAL
RUNTIME / GATEWAY / MCP / POLICY / MEMORY / OBSERVABILITY / EVALUATIONS
LIVE
Automation Portal

Etch Process Window
Factory Engineering

A single automation portal for senior AWS developers building etch process window test automation, MCP tool fabric, and governed photolithography drift orchestration. The portal connects runtime deployment, tool discovery, deterministic calculations, policy gates, memory summaries, telemetry, evaluations, and enterprise adoption patterns.

AGENTCORE RUNTIMEAGENTCORE GATEWAYSTRANDS AGENTSMCPSIGV4JSON-RPCAWS LAMBDAAMAZON BEDROCK
Runtime Entry Points
3
Gateway Tools
4
Specialists
4
Policy Checks
2
Evaluation Sets
5
Session State
ON

Enterprise Automation Control Plane

Built for experienced AWS developers who need architecture depth, deterministic test points, and audit-visible workflow boundaries.
AgentCore + Strands + MCP

Runtime Agent Workbench

Build synchronous, streaming, and large-payload AgentCore Runtime entrypoints backed by Strands Agents and Amazon Bedrock models. Validate payloads locally, run smoke tests, deploy with the starter toolkit, invoke with boto3, and clean up sessions.

app.pyapp_streaming.pyapp_large_payload.py

MCP Tool Fabric

Create canonical tool schemas, local streamable HTTP MCP tools, Lambda-backed Gateway targets, semantic search, direct JSON-RPC diagnostics, and Strands consumption over SigV4 transport.

tools/listtools/callsemantic search

Governed Multi-Agent Flow

Run an orchestrator with throughput, metrology, overlay, and etch-process specialists. Add request policy, response policy, bounded memory summaries, structured observability events, evaluation fixtures, and trace-aware response contracts.

policymemorytelemetryevaluations

Architecture Layers

  • Client portal sends request ID, user ID, trace ID, prompt, and optional data payload.
  • AgentCore Runtime hosts deterministic tools and Strands agent entrypoints.
  • AgentCore Gateway publishes Lambda-backed MCP tools with JWT ingress and IAM target access.
  • Governance layer coordinates policy, memory, observability, and evaluation evidence.

Adoption Notes

  • Use local validation before managed runtime deployment.
  • Keep schemas source-controlled and linted before Gateway registration.
  • Separate semantic discovery tests from direct tool execution tests.
  • Propagate request ID and trace ID across specialists, logs, and responses.

AgentCore Runtime Automation

Synchronous, streaming, and large-payload agent runtime workflow.
Local → Deploy → Invoke → Cleanup

Runtime Payload Contract

required: prompt optional: request_id, user_id, excel_data, image_data, metadata validation: reject missing prompt and unknown top-level fields session: runtimeSessionId reused for multi-turn workflow cleanup: stop_runtime_session after workflow completion

Deterministic Tool Pattern

yield_drift_bpu(fab_a_etch_rate, fab_b_etch_rate) overlay_control_count(wafer_count, control_ratio) scrap_impact_notional(lot_count, drift_move_bpu, queue_time) Use Python tools for repeatable calculations. Use Bedrock model for synthesis and limitations.

Gateway MCP Tool Fabric

Managed MCP endpoint with schema governance, semantic discovery, Lambda target logic, and raw JSON-RPC debugging.
4 tools registered

Gateway Flow

  1. Define canonical tool_schema.py.
  2. Run local FastMCP server over streamable HTTP.
  3. Verify tools/list and direct tool execution locally.
  4. Package Lambda target logic.
  5. Create AgentCore Gateway with MCP protocol and semantic search.
  6. Register Lambda target with inline tool schema.
  7. Run semantic search and direct JSON-RPC tools/call.
  8. Load Gateway tools into Strands agent over SigV4 transport.

Debug Payload

{ "jsonrpc": "2.0", "id": 201, "method": "tools/call", "params": { "name": "process_window_drift", "arguments": { "query": "Fab-A/Fab-B yield drift widening, overlay drift, and lot-flow stress" } } }

Governed Photolithography Drift Automation

Multi-agent orchestration with policy, memory, telemetry, evaluations, and response contracts.
orchestrator + 4 specialists

Policy Boundary

blocked request patterns: - execute equipment action - change equipment state - bypass approval - guaranteed yield improvement - hide scrap signals - evade controls - ignore process limits required response terms: - confirmation - invalidation - limitations

Telemetry Event Shape

{ "timestamp": "ISO-8601", "elapsed_ms": 1280, "event_type": "specialist_complete", "request_id": "hands-on-governed-001", "attributes": { "specialist": "metrology", "confidence": "medium" } }

Etch Process Window Automation Leaderboard

Original interactive leaderboard behavior retained and converted into process-window automation telemetry.
8 test cells
#
Owner
Index
Delta
Spark
Stability
PF
WR
Max Drift
Analysis

Automation Portal Runbook

Two-hour build path plus production-hardening gates for enterprise teams.
developer checklist

Promotion Gate

  • Payload schema validation passes.
  • Tool schema linter validates all Gateway tools.
  • Gateway tools/list smoke test returns registered schemas.
  • Semantic search evaluation returns expected tool in top results.
  • Policy unit tests pass for allowed and blocked prompts.
  • Structured telemetry includes request ID and trace ID.

Production Backlog

  • Move local memory abstraction to managed memory capability.
  • Replace simple policy checks with managed policy controls where appropriate.
  • Route structured telemetry to managed observability.
  • Convert JSON fixtures into managed evaluation datasets.
  • Add IAM roles per runtime environment, timeout budgets, retries, and fallback behavior.