Technical overview

From wafer map to defect pattern JSON in under 3 seconds.

Wafertune is an API service. You send wafer map data. We return structured defect pattern classifications.

Data flow

How your data moves through Wafertune

Supported inputs

Wafer map formats Wafertune accepts

No preprocessing layer needed — Wafertune normalizes format differences at the API boundary.

Format	Description	Required fields
STDF	Standard Test Data Format — WaferSort (WSR), EWS lot/wafer records. Most common in analog and power-device EWS flows.	WRR, WIR records; die pass/fail bin data
CSV die map	Binary or multi-bin CSV grid. Row-major order, X/Y die coordinates inferred from grid position or explicit columns.	Bin value per die cell; wafer diameter (optional)
PNG wafer map	Rendered wafer map image. Supports standard color conventions for pass/fail zones.	Min 300×300px; aspect ratio approximately 1:1
HDF5	Hierarchical data format. Common in research fab environments and inline inspection tool exports.	Dataset path; die coordinate arrays

Response schema

What comes back from the API

Every classification response is a structured JSON object. Fields are stable across model versions.

Response: POST /v1/classify

{
  "wafer_id": "LOT42_W03_EWS",
  "classification_time_ms": 2381,
  "pattern_classes": [
    {
      "class_id": "RING_EDGE_EXCL",     // Wafertune taxonomy ID
      "confidence": 0.94,              // 0.0 – 1.0
      "bbox": [0, 0, 300, 300],       // [x, y, w, h] in die units
      "process_origin_hint": "edge_bevel"
    },
    {
      "class_id": "CLUSTER_RANDOM",
      "confidence": 0.81,
      "bbox": [120, 95, 45, 52],
      "process_origin_hint": "particle_fallout"
    }
  ],
  "overall_yield_risk_score": 0.72,   // 0 = no risk, 1 = high risk
  "model_version": "wft-v1.4.2"
}

Integration paths

REST API direct vs Python SDK

REST API (any language)

Direct HTTP

Standard POST request, JSON body
Works with curl, requests, axios, fetch — any HTTP client
API key in Authorization header
Full response schema available immediately
Best for: existing pipeline integration, R, Julia, Perl, shell scripts

Python SDK (recommended)

wafertune-python

pip install wafertune
Handles STDF parsing, format detection, retry logic
Batch submission with async support
Typed response objects for IDE completion
Best for: new integrations, Jupyter notebooks, automation scripts

Performance

Latency and reliability

<2.5s P50 classification latency

<4s P99 classification latency

99.5% Uptime SLA (Production tier)

100 Max batch size per request

Retry behavior

The API returns HTTP 429 on rate limit and HTTP 503 on temporary capacity. Both are safe to retry with exponential backoff (suggested: 1s, 2s, 4s, 8s). The Python SDK handles retry automatically with configurable max_attempts.

Ready to see the API reference?

Full endpoint documentation, authentication guide, and error code reference.

See the API Reference Request Pilot Access