Semiconductor ERP + AI
AI for Semiconductor ERP: Wafer Lots, Yield Planning and Export Control
Short answer
In semiconductor companies the ERP carries planning, subcontract flows, inventory and finance, while the fab or test floor runs on MES and equipment systems. AI over the ERP answers lot genealogy, yield-adjusted supply and export control questions from your own data, running privately so process and customer information stays inside your network.
- ERP
- SAP S/4HANA, Oracle, Infor, Microsoft Dynamics 365
- Industries
- Semiconductor, Electronics, OSAT and Assembly Test
- Written for
- Supply Chain Director / ERP Manager
Semiconductor is a poor fit for generic ERP assumptions. A single wafer lot splits and merges, die yield varies by lot and bin, a finished device may pass through foundry, wafer sort, assembly and final test at different vendors, and lead times run for months. Fabs run on a dedicated MES with equipment integration over SECS/GEM, so the ERP usually does not hold in-fab detail. Fabless firms and integrated companies lean on the ERP for subcontract purchasing, inventory across the chain and customer commitments.
That makes the ERP the place where supply questions come to a head. How many good die will be available for a customer order in a quarter, given wafer starts, sort yield and assembly loss? Which lots are on hold at the OSAT, and which orders do they affect? Which products need an export license for a destination? The data to answer exists, but across ERP, MES exports and spreadsheets from partners.
Netray builds private AI on the ERP and the data around it. Process data, yields and customer roadmaps are among the most sensitive information in the sector, so models run on your hardware or in a sovereign cloud with no external calls. The AI reads and cross-checks; planning decisions and release authority stay with your teams.
This page explains where AI helps, where the MES is the right system, and what compliance questions matter, including export controls and customer quality requirements.
What usually gets in the way
The problems we hear most from supply chain director / erp manager teams running SAP S/4HANA.
Yield-adjusted supply is hard to see
Converting wafer starts into good die available for orders depends on lot yields, sort bins and assembly loss. These sit in several systems, so commitment dates are based on stale averages.
Lot genealogy crosses company boundaries
Wafer, die and packaged device lots are split, merged and moved between foundry, sort, assembly and test partners. Tracing a field return to wafer lot and process step requires partner data that arrives in different formats.
Subcontract visibility is thin
Work at OSATs and foundries is tracked through shipments, status files and email. The ERP often shows inventory at vendor but not what stage it has reached.
Export control and customer requirements add checks
Classification under export regulations, restricted destinations and customer-specific quality flow-downs such as automotive requirements apply to specific parts, and checks are manual.
Long lead times punish small errors
A wrong forecast or an unnoticed hold shows up months later, and planners spend time reconciling data instead of planning.
Where AI earns its place in SAP S/4HANA
Each use case names the ERP objects it reads or writes, so your ERP team can judge the integration effort before anyone commits budget.
Yield-adjusted availability
Compute good-die availability by product and quarter from wafer starts, lot yields and assembly loss factors, then compare against open orders.
Touches: Production and subcontract orders, BOM with yield factors, lot attributes, sales orders, planning runs
Outcome: Order promising is based on lot-level yield, not a single average, with assumptions displayed.
Wafer-to-device genealogy
Trace a packaged device lot back to wafer lots and forward to customer shipments, listing the partners and steps involved.
Touches: Lot master, subcontract movements, goods receipts, delivery documents, partner lot references
Outcome: Return investigations begin with a complete chain, with gaps highlighted where partner data is missing.
OSAT and foundry status consolidation
Combine ERP subcontract orders with partner status files to show where each lot is and which are late.
Touches: Subcontract purchase orders, vendor-stock inventory, shipping notices, partner status feeds
Outcome: Planners see one status list instead of opening several vendor portals.
Hold impact analysis
When a lot is placed on hold, list the orders, customers and downstream lots affected and the earliest alternative supply.
Touches: Lot status, open sales orders, planned orders, alternate lots
Outcome: Faster impact scoping that supports a decision by planning and quality.
Export control screening support
Flag orders where the product classification and ship-to country need license review, listing the supporting records.
Touches: Item classification fields, sales order ship-to, customer master, compliance holds
Outcome: Reviewers see a pre-screened list; license determinations stay with the trade compliance team.
Customer quality flow-down checks
Compare customer-specific requirements, such as automotive qualification or change notification rules, against the parts and lots shipped.
Touches: Customer-item records, change history, certificates, delivery records
Outcome: Missed flow-downs are found before shipment rather than at customer audit.
Last-time-buy and end-of-life planning
Estimate remaining demand against inventory and wafer bank for discontinued parts and list affected customers.
Touches: Item lifecycle fields, forecast, inventory by stage, customer demand
Outcome: Decision-ready summaries for last-time-buy offers, bounded by forecast reliability.
Reference architecture
Read-only retrieval across ERP planning and inventory data and partner or MES exports, with models hosted entirely inside your environment.
- 1
ERP connectors
Read-only APIs, views or a replica for orders, subcontract flows, lots and delivery data.
- 2
MES and partner data ingestion
Scheduled imports of MES yield summaries, test results and partner status files (CSV, EDI or API) into a governed store.
- 3
Role-aware retrieval
Customer, product and site limits applied so process and customer data are visible only to authorized staff.
- 4
Open-weight model serving
Models served on local GPUs or a sovereign cloud, with yield math done in queries and shown in the answer.
- 5
Evidence and logging
Answers cite records and assumptions, and queries are logged for audit.
Integration notes for your ERP team
- Do not expect the ERP to hold in-fab detail. Pull yield and test summaries from the MES or yield management system into a governed store.
- Model wafer, die and device lots with explicit parent-child lot relations. Without them genealogy answers are guesses.
- Partner status arrives as files or EDI messages, so normalize lot identifiers across partners during ingestion.
- Keep yield-adjusted calculations in SQL with versioned factors, and display assumptions with every answer.
- Map export classification fields at item level and keep them separate from the model's reasoning.
- Run read-only. Changes to planning parameters should flow through your normal planning process.
Deployment options
On-premise or air-gapped
Companies with defense or sensitive process data
Local GPU infrastructure with no internet access, with updates delivered by controlled transfer.
Sovereign or private cloud
Fabless firms with cloud ERP
Dedicated GPU instances in the region required by export or customer rules.
Plant-level and group-level split
Multi-site companies
Local inference near sensitive sites with a group index for planning questions.
Compliance and data control
How the architecture supports your obligations. Certification and accountability stay with your organisation; the design keeps the evidence straightforward.
US export controls (EAR) and ITAR where applicable
Hosting inside a controlled boundary helps the customer meet access restrictions. Classification and licensing remain with the trade compliance function.
Customer quality standards such as AEC-Q100 flow-downs and IATF 16949 supplier requirements
The AI cross-checks records against customer requirements for review, without certifying conformance.
RoHS, REACH and conflict minerals reporting
Retrieves material declarations and supplier data to support reporting prepared by your compliance team.
Confidentiality of process and customer data
Models run in your boundary with role controls, and no data is used to train external models.
How an engagement runs
Phase 1 . 2 to 3 weeks
Flow and data mapping
- -Supply chain and lot model
- -ERP, MES and partner data inventory
- -Security and export boundary requirements
Phase 2 . 5 to 6 weeks
Pilot on one product line
- -Genealogy and status queries
- -Yield-adjusted availability prototype
- -Comparison with planner results
Phase 3 . 4 to 6 weeks
Extend to compliance checks
- -Export screening support
- -Customer flow-down checks
- -Role mapping
Phase 4 . Ongoing
Operate
- -Accuracy and usage monitoring
- -Partner onboarding
- -Factor updates under change control
Questions to ask any vendor, including us
A short list that separates real SAP S/4HANA AI work from a chatbot demo.
- Which data comes from the ERP and which from the MES or partners?
- How are lot splits and merges represented in genealogy answers?
- Where are yield factors stored and who owns them?
- Can the system run fully offline or in a controlled boundary?
- How is partner data normalized and validated?
- What export or licensing decisions does the tool avoid making?
- How are customer-confidential roadmaps separated by user?
Frequently asked questions
Does the ERP replace a fab MES?
No. Fabs rely on MES with equipment integration for in-process control. The ERP covers planning, procurement, subcontract flows, inventory and finance. AI over the ERP, combined with MES summaries, supports planning and traceability questions, not process control.
Can AI predict yield?
Predictive yield modelling is a data science project using process and test data, usually in dedicated systems. What we do is make existing yield data visible alongside orders and inventory, with calculations done transparently in queries.
How does it help with export control?
It pre-screens orders using classification and destination fields and lists the supporting records. Classification and license decisions remain with trade compliance.
Can it work air-gapped?
Yes. Open-weight models can be served on local GPUs with no internet access. Model updates and data refreshes are delivered through controlled transfer procedures.
Which ERPs fit?
SAP, Oracle, Infor and Dynamics 365 are common in the sector. We connect through APIs, views or replicas and confirm how subcontract flows and lots are modelled before promising answers.
What about OSAT partner data?
Partner status and test data usually arrive as files, EDI messages or portal exports. We ingest and normalize them into a governed store so genealogy and status answers cover the full chain.
Is this worth it for a small fabless company?
Often the simpler route is better: clean lot relations and a few ERP reports. AI adds more once there are multiple partners, many products and frequent hold or late-lot questions.
Related guides
AI for ERP in Taiwan: Electronics, Semiconductor Supply Chain, and On-Prem AI
AI for SAP, Oracle, Infor LN and Digiwin ERPs in Taiwan electronics and semiconductor manufacturing: customer-NDA-safe AI, PDPA, and on-prem deployment design.
Buyer guide: cloud AI vs private AIFedRAMP / GCC High AI vs On-Prem AI for Your ERP: An Honest Comparison
An honest CISO comparison of FedRAMP High or GCC High AI copilots versus on-prem private LLMs for ERP data: what each authorization actually covers, and when each fits.
RAG + SQL + permissionsA Private LLM Grounded on Your ERP Data
How a private LLM answers questions on your ERP data: RAG plus text-to-SQL, role-based permissions inherited from the ERP, and where each fits.
On-prem AI, any ERP, A&DOn-Prem AI for ERP in Aerospace, Defense, and Electronics Manufacturing
A hub guide to on-prem AI across SAP, Infor LN, Costpoint, IFS, and Oracle EBS for aerospace, defense, and electronics manufacturers under ITAR, CMMC, and AS9100.
Agents + approval gatesAI Agents for ERP, Running On-Prem
A practical guide to on-prem AI agents for ERP: what they can safely automate, where human approval belongs, and how to design the guardrails.
Digital thread + AIAI Across PLM and ERP: Teamcenter, Windchill, and Your ERP
AI that reads Teamcenter or Windchill alongside your ERP to trace engineering change impact, BOM sync gaps, and where-used questions, on-prem.
Plan it with numbers
Air-Gapped AI Readiness Assessment
A 10-question assessment that scores how prepared your organization is to deploy and operate LLMs inside an air-gapped or classified enclave.
Free ToolGPU Sizing Calculator for LLM Inference
Work out how many GPUs you need to serve a given open-weight model to your user base, based on memory footprint and token throughput.
Free ToolOn-Prem vs Cloud AI Cost Comparison Calculator
Compare your projected cloud LLM API bill against an amortized on-prem GPU deployment and find the breakeven month.
GuideSyteLine SDK and API Reference Guide for Developers
Complete SyteLine SDK and API reference. Mongoose namespaces, IDO runtime APIs, web service endpoints, and developer toolkit for CloudSuite Industrial.
GuideSyteLine IDO Security and Permissions Setup
Configure SyteLine IDO-level security and role-based permissions. Control read, insert, update, and delete access per IDO, property, and user group in CloudSuite Industrial.
GuideSyteLine Mongoose Framework Architecture Deep Dive
Deep dive into SyteLine Mongoose framework architecture. IDO runtime, request pipeline, session management, and extensibility patterns for developers.
Talk it through with an engineer who knows SAP S/4HANA
Bring one real question your team cannot answer from the ERP today. We will map the data path, the model, and where it runs, and tell you honestly if AI is the wrong tool for it.