aPlanner - AI-Powered Production Planning
aPlanner - AI 기반 생산계획 솔루션
Plan Smarter. Produce Faster.
An AI production-planning engine built for semiconductor complexity: yield-aware, constraint-based and scenario-driven. Four AI modules, one planning engine, designed to cut master-planning cycles from days to hours and to run inside the fab network.
Design targets from simulation on synthetic fab data; no customer deployments yet.
The world’s most advanced fabs plan in spreadsheets
When a machine goes down or a hyperscaler doubles its order, a planner needs a new plan in hours. Today it takes days, in Excel, with numbers nobody fully trusts. Industry estimates put the cost at US$100M+ a year per large fab in over-/under-production, missed demand and excess inventory.
Spreadsheet-based planning
3–5 days per master-planning cycle; plans are outdated before they are finished
Inaccurate demand forecasts
55–65% forecast accuracy with Excel; AI / data-center orders swing 30%+ in a quarter
Manual constraint management
~70% of schedules need manual fixes; knowledge lives in planners’ heads
No scenario analysis
A what-if takes days, so planners test 1–2 options and decide by gut feel
Disconnected systems
Data siloed in MES, ERP and spreadsheets; management sees yesterday’s numbers
Reactive firefighting
~70% of planner time spent gathering data and handling exceptions, not strategy
Why it is so hard in semiconductors
AI-powered planning in hours, not days
aPlanner replaces spreadsheets and gut feel with AI that forecasts demand, optimizes schedules across all constraints simultaneously, and lets planners test every decision before they make it. Each module is a replacement for a spreadsheet, not a dashboard on top of one.
Target performance — MVP under validation (2026)
| Metric | Excel-based planning today | aPlanner (target) | Target improvement |
|---|---|---|---|
| Planning cycle | 3–5 days | Target: < 30 min full-fab schedule | Target 80% faster |
| Forecast accuracy | 55–65% (Excel) | Target: 85–90% | Target +20–30 pts |
| Scenario analysis | Days per scenario | Target: minutes per scenario | Target 95% faster |
| Schedule feasibility | ~70% (needs manual fixes) | Target: 99%+ (all constraints) | Target near-perfect |
| Planner time on strategy | ~20% | Target: 60% | Target 3x more |
| Data freshness | Yesterday’s data | Target: one live source of truth | Always current |
Design targets from simulation on synthetic fab data; no customer deployments yet. Baseline figures are industry estimates for Excel-based fab planning.
Four AI modules, one planning engine
Forecasting tells the fab what the market will want; scheduling turns that into a feasible, yield-aware plan; simulation lets planners test every decision before they make it; analytics keeps everyone on the same live numbers.
AI Demand Forecasting
Ensemble ML (LSTM, Prophet, XGBoost, Transformer) forecasting demand per bit-density tier / SKU, 1 day to 24 months, with confidence intervals. The best model per segment is chosen automatically and retrained as data arrives.
Excel forecasts reach 55–65% accuracy, which means over- or under-production worth US$20–50M a year per large fab. AI / data-center orders swing 30%+ in a quarter.
Target 85–90% accuracy (+20–30 pts); less inventory in the wrong place; demand sensing that catches hyperscaler swings weeks earlier.
Intelligent Scheduling
Constraint-based optimizer across tools, materials, labor, maintenance and changeovers, yield-aware over 50–100+ wafer steps, with automatic re-planning on disruptions.
3–5-day manual cycles; ~70% of schedules need manual fixes; plans are outdated before they are finished; knowledge stays in planners’ heads.
Target full-fab schedule in under 30 minutes (80% faster); target 99%+ feasible plans; bin-sort allocation that maximizes revenue per wafer.
Scenario Simulation
Monte-Carlo what-if engine with templates (tool down, +20% hyperscaler order, yield −10%, node migration), side-by-side KPI comparison and an AI-recommended plan.
A scenario takes days in Excel, so planners test 1–2 options and decide by gut feel: an estimated US$15–25M a year of sub-optimal decisions per large fab.
Target minutes per scenario (95% faster); unlimited scenarios; quantified risk; decisions the whole team can defend.
Real-Time Analytics
Live planning KPIs (adherence, OTD, WIP, utilization), anomaly detection and explainable recommendations, with Korean / English planner and executive views.
Data siloed in MES, ERP and spreadsheets; planners spend ~70% of their time gathering it; management sees yesterday’s numbers.
One always-current source of truth; target planner time on strategy from 20% to 60%; every recommendation auditable.
Yield-aware, scenario-driven scheduling for multi-step wafer processes
The novelty is in combining three things that today live in three different heads: yield, constraints and scenarios.
Yield-aware constraint scheduling
The optimizer takes live yield and bin distributions per process step as inputs, so wafer starts, expected good die and bin mix are planned together across 50–100+ steps, instead of scheduling starts and hoping the yield follows.
Plans that survive ±10–15% weekly yield swings; fewer wafer starts for the same output (target 5–15% yield-adjusted gain); bin allocation that maximizes revenue per wafer. No incumbent APS models yield inside the schedule.
Node-migration planning
Models learning curves, ramp rates and capacity shifts when a line moves 176L → 236L → 300L+, planning old and new nodes as one system with shared tools and demand.
Smooth transitions without demand gaps; customer commitments protected during the ramp; capacity decisions visible months earlier. Generic tools treat a new node as just a new product.
Explainable re-planning
Every schedule change carries the triggering event, the constraints it hit and the KPI delta; the planner approves, edits or rejects in Korean or English.
Adoption in conservative fab planning cultures; auditability for management and quality; planners keep control while tribal knowledge moves from heads into the system.
Where aPlanner is today
Not commercialized; no revenue; no customers yet. Working prototypes of scheduling, simulation and analytics on synthetic fab data; forecasting engine prototyped, dedicated view in the December 2026 build. All are being validated ahead of the 2027 Korean pilot program.
Phase 1 — Foundation
Apr–Jun 2026
Phase 2 — Concept
Jul–Sep 2026
Phase 3 — Validate
Oct–Dec 2026
Phase 4 — Launch prep
Jan–Mar 2027
Legacy APS is generic, slow and bolted-on AI
Incumbent planning suites were not designed for yield curves and node transitions, and the AI-first vendors are built for consumer supply chains. aPlanner column shows design targets.
| SAP IBP / Oracle APS | Kinaxis / o9 / Blue Yonder | aPlanner (target) | |
|---|---|---|---|
| AI | Bolt-on | Partial / generic | AI-native |
| Semiconductor depth | Generic | Automotive / supply-chain focus | NAND- / fab-specific models |
| Deployment | 12–18 months | 4–12 months | Target 8–12 weeks |
| Scenario speed | Hours–days | Hours | Target minutes |
| Entry price | US$400–500K+/yr | US$250–300K+/yr | KRW 120–300M/yr (≈ US$100–250K) |
| Korean-fab data residency | Cloud-first | Cloud-first | On-prem / private cloud option |
Built for Korean fabs
Korean fabs ask three questions first: where is my data, does it talk to my MES, and who answers the phone in Korean? aPlanner is designed to answer all three, with KRW contracts from a Korean company (Amoza Korea Co., Ltd., incorporating 2027).
Data stays in the fab
On-premise or private cloud inside the fab network, with an air-gapped option. Read-only connectors first; no production data leaves the site. SaaS on AWS Seoul for suppliers and SMEs.
Korean systems landscape
Pre-built connectors for in-house MES, SAP ERP, yield / bin databases and planner spreadsheets (in development). Write-back of approved plans only after sign-off. Korean SI / MES partners for delivery.
Korean users, Korean support
Korean UI, reports and planner-approval workflow. Korean hires in 2027 (planning domain, MES integration, business development). Founder and support within an hour of the Yongin–Pyeongtaek cluster.
Security & governance
Role-based access, SSO, full audit trail; AI recommends, people decide. Roadmap: SOC 2 Type II (2027–28), ISO 27001 (2028), K-ISMS (planned). No certification is held today.
Editions (Korea, annual, KRW)
| Edition | Annual | For |
|---|---|---|
| Starter | KRW 120–300M / yr (≈ US$100–250K) | Single site, core AI (forecasting, scheduling, basic scenarios) |
| Professional | KRW 400M–1B / yr | Multi-site, MES + ERP connectors, Monte-Carlo simulation |
| Enterprise | KRW 1B–2.5B+ / yr | Global optimization, custom models, 24×7 support |
Services
Indicative pricing for the 2027 Korea pilot program and first subscriptions. See pricing
From readiness assessment to subscription
Land with one line, prove the plan against yours, then expand. Designed for Korean memory fabs, tier-1/2 suppliers and OSATs, and high-mix electronics, display and battery plants.
Paid Planning Readiness Assessment
Data inventory (MES, ERP, yield / bin DB, spreadsheets), planning-process mapping and a quantified value hypothesis for your site. KRW 50–100M, 3–4 weeks.
8–12-week pilot on one production line
aPlanner deployed on-prem or in a private cloud against one line’s real data, read-only connectors first. Target KPIs agreed up front and measured side by side with the current plan. KRW 75–150M.
Starter subscription
Convert the pilot into a Starter subscription for the site, then expand to multi-site Professional and, as the MAKE roadmap matures, aMonitor.
aPlanner is the starting point, not the destination
Planning leads to monitoring (aMonitor), and planning plus monitoring on one AI platform becomes a complete Smart Factory. The products below are roadmap; today we build aPlanner.
Become a design partner in 2027
We are looking for a small number of Korean fabs and suppliers to pilot aPlanner on one production line: paid readiness assessment, 8–12-week pilot, then a Starter subscription.