APS for Fabless Semiconductor Manufacturing: Planning & Scheduling

Last Update: October 16, 2026

Advanced planning and scheduling (APS) for fabless semiconductor manufacturing coordinates wafer starts, wafer and die inventory, assembly and test capacity, and customer orders across external foundries and outsourced semiconductor assembly and test (OSAT) providers.

A fabless semiconductor company designs and sells chips but outsources their manufacture. Its planners still own the outcome. They decide how many wafers to start at each foundry, where to hold inventory between stages, which OSAT builds which product, and which customer gets supply when capacity is short. This guide explains how that planning works, walks through a worked example from demand to wafer starts, and sets out what to test in APS software. Eyelit Technologies provides advanced planning and scheduling software for manufacturers through Eyelit APS.

Key terms

• Fabless semiconductor manufacturing: chip design and sales, with manufacturing outsourced to foundries and other production partners.
• OSAT: an outsourced semiconductor assembly and test provider.
• APS (advanced planning and scheduling): software that coordinates constrained supply, capacity, inventory, and production plans across manufacturing stages.

How do fabless semiconductor companies plan and schedule across outsourced manufacturing?

Fabless semiconductor companies plan across outsourced manufacturing by treating the supply chain as a series of stages, each with its own buffer, planning horizon, and partner, and by planning all of them in one model. Five practices do most of the work:

1. Commit wafer starts against forecast. Wafer fabrication takes months, so wafer starts are committed before most customer orders arrive. How far ahead depends on the foundry’s lead time, yield variability, and product mix.
2. Hold inventory in banks between stages. Finished wafers wait in a wafer bank, tested die in a die bank, and packaged parts in finished goods. Each bank decouples adjacent stages with different lead times, capacity constraints, or sources of variability.
3. Plan the reverse bill of materials. One wafer becomes many die of different grades, so planning works from the products needed back to the wafers required, using expected yields and bin splits.
4. Carry customer requirements through the plan. Customer qualifications, such as approved sites, wire type, or wafer revision, decide which inventory can serve which order.
5. Replan as partner data arrives. How often to replan depends on how often foundries and OSATs report work-in-progress and yields, and on how quickly demand or capacity changes need to show in the plan.

For a broader view of supply chain planning for fabless companies, see how fabless semiconductor manufacturers reduce cycle times with planning.

What makes fabless semiconductor planning different?

Fabless planning differs from planning inside one factory because the company plans stages it does not run, with partners that report on their own schedules. The table below shows one illustrative supply chain. Who runs probe and final test varies: some fabless companies keep them in-house, others outsource them.

Stage	Often run by	Inventory held after it	What the planner decides
Wafer fabrication	Foundry	Wafer bank	How many wafers to start, at which foundry, and on which process
Wafer probe	The fabless company or a test partner	Die bank	Which wafers to probe, and how test results change die available by grade
Assembly	OSAT	Usually none, as parts move on to final test	Which OSAT builds which package, and when
Final test	The fabless company or an OSAT	Finished-goods inventory, at the applicable stocking location	Test capacity by tester and handler setup, and how bins split into products
Distribution	Distributors	Distributor inventory, owned by the distributor once sold	How much to ship to each channel, and how much sits unsold at distributors

Three things make this harder than single-plant planning:

• Different horizons. Wafer fabrication can involve long planning horizons, while downstream assembly and test have their own lead times and capacity constraints. The plan must accommodate both.
• Many partners. Several foundries and OSATs can be qualified for the same product, each with its own cost, lead time, and minimum commitment.
• Late data. Partner work-in-progress and yield data often arrive as periodic files rather than live feeds. The plan has to absorb each update without being rebuilt.

How does APS handle binning, the inverted BOM, and customer qualifications?

APS handles these by planning from finished products back to wafers, and by carrying product and customer attributes through every stage. Three ideas sit behind it:

• Binning. Test sorts die into bins by speed, power, or quality grade. The same wafer produces several bins.
• Bill of Grades. A Bill of Grades describes how a wafer is expected to split across bins. Actual test results replace the expected split once a lot is probed.
• Inverted BOM. In most manufacturing, many parts make one product. In semiconductors, one wafer makes many products. This is the inverted bill of materials (BOM), and planning has to work it backward, from products to wafers.

Customer requirements add a second layer. One part number can serve several customers, but each customer may only accept material from certain sites, with a certain wire type, or with a given wafer revision or date code. Ignore these and inventory that looks available cannot ship.

Eyelit APS supports attribute-based planning for semiconductor manufacturing — customer qualifications such as Au/Cu wire, date codes, and wafer revisions carried through pegging — with binning, Bill of Grades, and inverted-BOM support from wafer to die to finished goods.

More on Eyelit’s semiconductor planning is on the Eyelit semiconductor page.

Worked example: from finished-product demand to wafer starts

This example uses made-up figures to show how binning, yields, and qualifications interact. It is illustrative only. It is not output from Eyelit APS and not data from any customer.

Assumption	Value
Gross die per wafer	500
Probe yield	80%, giving 400 good die per wafer
Bill of Grades (share of good die)	Grade A 25%, Grade B 50%, Grade C 25%
Assembly and final test yield	95%
Finished-product demand for the period	Grade A 10,000, Grade B 12,000, Grade C 4,000

Step 1: work back from demand to die. At 95% assembly and test yield, the demand above needs 10,527 A die, 12,632 B die, and 4,211 C die.

Step 2: work back from die to wafers.

Grade	Die per wafer	Die needed	Wafers required if this grade sets the plan
A	100	10,527	106
B	200	12,632	64
C	100	4,211	43

The three wafer figures do not add up, because every wafer produces all three grades. The wafer start is the largest of the three, under these yields and this bin split.

Grade A sets the wafer start at 106 wafers. Those wafers also produce 21,200 B die and 10,600 C die, well above demand. The surplus goes to the die bank, where it can serve later orders, or the plan can shift the demand mix toward B and C.

Step 3: change one assumption. If Grade A falls from 25% to 20% of good die, each wafer gives 80 A die instead of 100, and the wafer start rises from 106 to 132. That is approximately 25% more wafer starts under these assumptions, from a five-point change in the bin split. Expected bin splits need updating as real test results come in for exactly this reason.

Step 4: add a customer qualification. Suppose 3,000 of the Grade A units are for a customer that only accepts wafer revision 2. At 95% assembly and test yield, that order needs about 3,158 revision 2 Grade A die. The die bank holds 2,000 Grade A die, all from revision 1, and none of them can serve this order. The planner has to source enough revision 2 die to cover it, or flag the shortfall and revise the promise date. The revision 1 die stay available for other customers.

Step 5: add a capacity limit. If the main OSAT can assemble only 8,000 Grade A units in the period, the remainder has to move to an alternate OSAT qualified for that package, or to a later period. A good plan minimizes disruption to existing commitments, subject to capacity, customer priorities, and the planning constraints that have been configured.

How should fabless companies position inventory across wafer, die, and finished goods banks?

Fabless companies should evaluate inventory positioning across wafer, die, and finished-goods stages according to demand uncertainty, yield variability, downstream lead times, and customer service requirements. Earlier-stage inventory can preserve product-mix flexibility, while finished-goods inventory can support faster fulfillment where demand is sufficiently predictable.

Each bank needs its own floor and ceiling. The plan also has to show inventory building up ahead of a ramp and drawing down after it, without breaching either limit.

Eyelit APS models inventory as hard or soft constraints, including safety-stock floors, storage ceilings, downstream coverage rules, and build-up and build-down.

How do fabless companies promise orders when capacity is short?

Fabless companies promise orders by checking both inventory already in the pipeline and capacity still open at partners:

• Available-to-Promise (ATP): commits orders against existing or planned inventory in the banks.
• Capable-to-Promise (CTP): commits orders against foundry and OSAT capacity and materials not yet committed.

Both have to respect customer qualifications, since a promise built on inventory or capacity the customer cannot accept will be missed. When wafer capacity is tight, promising also turns into allocation. The planner decides which customers get the supply, and the promise system has to keep to those decisions.

Eyelit APS delivers Available-to-Promise and Capable-to-Promise in both batch and real-time modes, with the real-time module connecting to the order management system.

Why does the distributor channel matter for fabless demand planning?

The distributor channel matters because sales into distributors (sell-in) can run ahead of sales from distributors to end customers (sell-through). When that gap grows, inventory builds up at distributors, future orders fall, and a forecast based on sell-in alone overstates demand. Sell-through gives an earlier signal of real demand, and that signal feeds back into wafer starts months ahead.

Eyelit supports distributor-channel sell-in and sell-through inventory management.

How should the plan respond when yields or partner schedules change?

The plan should change as little as the new data allows. A low-yield lot, a late OSAT shipment, or a foundry capacity cut affects some orders, not all of them. Replanning with memory means the planning system aims to keep existing commitments where it can and to limit changes to the orders affected, within limits the planners set.

Eyelit APS replans with memory: when demand or supply changes, it minimizes disruption to the existing schedule, with the degree of allowed change configurable at any level or attribute.

Where the fabless company runs its own probe or final test, those sites can feed live status straight into planning. For the execution side of probe and test, see our guide to MES for semiconductor manufacturing.

Eyelit APS is advanced planning and scheduling software that runs on the same data model as Eyelit MES, so real-time WIP, downtime, and shop-floor changes feed updated schedules back to execution.

What should fabless companies test when evaluating APS software?

Test APS software on a week of your own data: real forecasts, foundry and OSAT work-in-progress files, bin splits, and customer qualifications.

What to evaluate	What to ask the vendor to show	What a good result looks like
Inverted BOM and binning	Wafer starts calculated back from finished-product demand, using your Bill of Grades	Wafer start quantities you can check by hand, with surplus die by grade shown
Bin split changes	The same plan rerun with a lower share of your top grade	The change in wafer starts shown, with the reason traced to the bin split
Customer qualifications	An order that can only ship from qualified sites or a given wafer revision	Ineligible inventory excluded from that order, and qualified supply pegged to it
Inventory banks	Floors and ceilings at wafer bank, die bank, and finished goods, through a ramp	No breaches, or each breach flagged with its cause
Partner capacity	Foundry and OSAT capacity limits, with an alternate qualified partner	Overflow moved only to qualified partners, at their stated capacity
Order promising	A new order promised against bank inventory and partner capacity	A promise date that respects qualifications, in batch and in real time
Replanning	A low-yield lot or a late OSAT shipment	Only the affected orders move, and you can see which ones and by how much
Distributor channel	Sell-in and sell-through compared by product	Distributor inventory shown by product, with the gap between sell-in and sell-through

For the general APS evaluation framework, see how to evaluate APS software.

Eyelit APS for fabless semiconductor companies

To evaluate Eyelit APS for fabless planning, bring a week of your own data: forecasts, foundry and OSAT work-in-progress files, your Bill of Grades, and the customer qualifications that limit which supply can serve which order. Then run the tests in the table above on it.

Each capability is covered earlier in this guide: attribute-based planning, binning, and the inverted BOM under customer qualifications; inventory constraints across banks; ATP and CTP; distributor sell-in and sell-through; and replanning with memory. Eyelit’s semiconductor capabilities are also described on the Eyelit semiconductor page.

Eyelit’s solver and optimization technology is patented (US8509925B2, US12229700B2).

For a side-by-side view of ten APS vendors, see our APS platform comparison.

Planning across foundries and OSATs? Bring a week of your forecasts, partner WIP, and bin splits, and see how Eyelit APS plans them. Eyelit for semiconductor

Frequently asked questions

What is a die bank?

A die bank is inventory held as tested die, after wafer probe and before assembly. Because a die can still become several finished products, a die bank lets fabless companies respond to demand changes faster than new wafer starts allow, without committing to a final package.

What is a Bill of Grades?

A Bill of Grades describes how a wafer is expected to split into die of different grades or speeds after test. Planners use it to calculate how many wafers to start for a given mix of finished products, then replace the expected split with actual bin results once lots are tested.

What is the difference between sell-in and sell-through?

Sell-in is what a semiconductor company sells to its distributors. Sell-through is what those distributors sell on to end customers. When sell-in runs ahead of sell-through, inventory builds up in the channel, so tracking both gives a more reliable view of real demand.

This guide is published by Eyelit Technologies.