Custom Power Management ICs: L1, L2 and L3 from Specification to Production
Stop Settling for Off-the-Shelf Part Numbers
STST's full-cycle custom power IC service: from L1 parameter customization to L3 full customization, a path you can scale by investment
Every hardware engineer knows this moment: the design is drawn, the BOM is pulled, and the power IC turns out to be one output-voltage step off, half a percentage point short on accuracy, and 0.4 mm taller than the board's height limit. Changing the circuit adds external parts, changing the mechanics means waiting for tooling, and changing the part number means redoing certification. So the usual verdict is: never mind, let's live with it for now.
But the cost of settling doesn't disappear. It just moves from the BOM unit price to somewhere else: into those few external components, one round of rework, a stretched lead time, the next design revision. This article sets out to make one thing clear: when a standard part no longer fits, what the custom path actually looks like, how long it takes, what it costs, what it requires from the customer, and which parts of the work STST can carry for you.
Lead time for some power devices after stretching from 8–12 weeks; higher still for automotive grades
Industry reference cycle for an all-new custom PMIC from kickoff to production
Why Talking About Customization Now Is Worth More Than Talking About Selection
For the past decade, selecting a power IC meant opening a selection table, filtering by input voltage, output current, package, and quiescent power, and picking the "closest" part. The method rests on three assumptions: ample supply, stable lead times, and prices that fall every year. Over the last two years, those assumptions have been rewritten.
The first change is lead time. Analog and power device lead times have stretched from the usual 8–12 weeks to more than 30 weeks, with automotive IGBTs and silicon carbide devices quoted at over 40 weeks (36Kr). 8-inch BCD lines remain fully loaded, and foundry quotes have risen 5%–15% (Sohu). When being able to buy a part at all becomes a variable, the "in stock" column of a selection table carries an entirely different weight.
The second change is that substitution has entered deep water. China's analog chip market is expected to reach RMB 245.1 billion in 2026 (China Commercial Industry Research Institute), but the hard part to replace isn't consumer devices. It is high-reliability applications in industry and automotive, because there a matching pinout isn't enough: parameters, certification, reliability, and supply cycle all have to be realigned together. The industry is already discussing "co-optimized" custom foundry work, where end-use requirements drive the design rather than standard parts being made to fit (East Money).
The third change is that the threshold for customization itself is being redefined. "Custom chip" used to mean building a new chip from scratch by default. An all-new custom PMIC commonly takes 15–24 months from kickoff to production (AnySilicon), which is unacceptable for most end-product programs. But that premise needs unpacking: the 15–24-month baseline assumes architecture definition, process development, and IP validation all start from zero. Starting from a verified process platform and IP library gives a completely different time structure. So customization that actually works is never all-or-nothing. It must be tiered by investment and built on a verified platform: small changes take a short, low-threshold route, and only large changes take the full process.
A standard part answers "is there one?" Customization answers "does it fit?"
And "does it fit?" has never been a yes-or-no question. It is a choice between levels.
Why Standard Parts Fall Short: Three Hard Reasons
"Falling short" rarely means poor performance. More often it means being just a little off, and on the board, that little can cost a lot.
Certainty of Lead Time and Supply
Standard-part lead times are set by the original manufacturer's capacity and channel inventory. End-product makers sit at the end of that chain with almost no bargaining power. Lifecycle risk makes it worse: once a part enters end-of-life (EOL), the alternatives are either a PCB change or recertification, and both take time.
Customization changes the nature of the relationship, from fighting for inventory to production scheduled by contract. Capacity commitments, change-notice periods, and last-time-buy windows, terms once found only in major-account agreements, come standard in a custom program.
Parameter Mismatch: A Small Gap Costs a Lot
This is the most common case and the most easily underestimated. A few real design sticking points:
- Output voltage options: The system needs 3.6 V, but the standard part has only fixed 3.3 V and 5 V options. An adjustable part can reach 3.6 V, but often can't deliver the combination of fixed high accuracy, microamp quiescent current, and a small package.
- Accuracy: The product requires ±1%, the standard part offers ±2%, and closing that 1% gap takes calibration or screening.
- Quiescent power: A battery product needs microamp-level standby, the standard part draws milliamps, and battery life takes the direct hit.
- Package height: The module height limit is 0.8 mm, the standard package is 1.0 mm, and the only option is to change the mechanics.
- Protection thresholds: Overvoltage, undervoltage, and current-limit trip points don't match the system's operating window, causing either nuisance trips or no protection at all.
Every "just a little off" ends up converted into external components, PCB changes, revalidation, and recertification. None of these costs appear in the BOM unit price, but all of them are real.
Differentiation and System Cost Structure
The value of a standard part is that it is generic, and so is its price: it must leave margin for every customer, so external compensation, protection, and filtering are all left for you to build. Customization can absorb those external functions into the chip: what took 3–5 discrete devices becomes a few blocks inside one IC.
For an end-product maker, this isn't about saving a few cents per unit. BOM lines, board area, placement stations, and test steps all shrink together. The change in system cost is often an order of magnitude larger than the change in device unit price.
L1 / L2 / L3: Lead Time and Cost in One Table
Once you decide to customize, the first question isn't "how soon can you deliver?" but "which level do we need?" Choose too high and you pay for investment you don't need; choose too low and the project comes back for rework during validation.
| Level | What the Customer Changes | What STST Does | Typical Lead Time | Cost Structure | Best Fit |
|---|---|---|---|---|---|
| L1Parameter | Output-voltage option, accuracy grade, protection thresholds, screening requirements | Adjusts on existing silicon: internal trim, test specification, and screening conditions; no new mask set | 4–8 weeks | No NRE; packaging and test-related costs are borne by the customer | One parameter step off, everything else fits |
| L2Derivative | Pin definition, package type, addition or removal of some functional blocks | Derivative design on an established IP platform, reusing verified blocks and processes | 8–16 weeks | Modest NRE + minimum order quantity (MOQ) | Pin-to-Pin or a specific package is required |
| L3Full custom | All-new specification and architecture; verified BCD platforms reused first on the process side | Specification definition, design, tapeout, packaging, and reliability validation on a verified BCD process platform and reusable IP library | 6–12 months | NRE + masks + package tooling + initial-order commitment | New system requirements that standard parts can't cover |
L3's 6–12 months doesn't contradict the industry's 15–24 months from Section 01. The difference is the starting point. The industry baseline usually assumes architecture definition, process development, and IP validation are all built from zero. STST's L3 is built on a verified BCD process platform and reusable IP library, which removes platform creation, process development, and first-time IP validation, while design, tapeout, packaging, and reliability validation remain. When an all-new process or architecture is needed, the cycle returns to the longer range. This is made clear during the feasibility assessment, not left until the quotation.
So STST's default advice is clear: don't use L2 if L1 solves the problem, and don't use L3 if L2 is enough. This isn't conservatism. It puts your money into the level that actually solves the problem.
A real example: STST's current LDO product line supports custom output voltage in 0.1 V steps and custom ±1% accuracy. Both sit at L1, so the customer pays no NRE and waits for no tapeout. The bind of "only 3.3 V and 5 V" can become "exactly 3.6 V" within 4–8 weeks.
STST's Custom Service at a Glance: Ten Steps from Requirement to Production
Custom projects rarely fail on technology. Most fail because requirements weren't clarified up front, terms weren't fixed in the middle, and responsibilities weren't drawn at the end. STST breaks the process into ten nodes in three groups: clarify requirements before approval, turn the specification into something verifiable during design and validation, and lock down later uncertainty through delivery assurance.
Group 1 · Before Approval: Clarify the Requirement
- 01Requirement reviewAsk six questions before quoting (see the checklist below). Without complete answers, any quote is only an estimate.
- 02Level selection and feasibility assessmentDetermine whether the need sits at L1, L2, or L3, and assess BCD process fit, reusable IP blocks, die area and cost, package feasibility, and target reliability grade (industrial or automotive AEC-Q100). The conclusion states plainly what can be done, what can't, and what must change to make it possible.
Group 2 · Design and Validation: Turn the Specification into Something Verifiable
- 01Specification draft (draft datasheet)Covers absolute maximum ratings, recommended operating conditions, electrical characteristics, protection and transient behavior, and acceptance criteria. This draft is the basis for all later validation.
- 02Package and pin planPin-to-Pin compatibility comes first so the customer needn't change the PCB. If a new package is truly required, tooling investment and lead time are included in the plan.
- 03Reference design and evaluation board (EVB)Schematic, BOM, layout guide, and measured waveforms, delivered for direct on-board validation by the customer.
- 04Measured validation reportMeasured on STST's own EVB, with the test board, test points, instruments, and temperature conditions stated. Datasheet typical values, competitor data, and simulation results are never substituted for measurements.
- 05Reliability validationItems run according to the target grade; automotive programs are scheduled against the AEC-Q100 grade requirements.
Group 3 · Delivery Assurance: Lock Down the Uncertainty
- 01Project plan and milestonesSpecification freeze → design completion → tapeout → first-silicon validation → engineering samples → reliability validation → pilot production → production release, with deliverables and customer confirmations stated at each node (L1 projects skip tapeout and first-silicon validation and go directly to engineering samples and measurement).
- 02Commercial planNRE, masks, package tooling, samples, volume pricing, MOQ, sample policy, capacity support, and quotation validity, itemized one by one with nothing left "to be discussed separately."
- 03Risk and responsibility allocationRespin responsibility and rounds, Product Change Notification (PCN) and End-of-Life (EOL) notice periods, Last-Time-Buy (LTB) window, IP ownership, and confidentiality terms.
The Custom Workflow in One Chart
Eight milestones, each with items the customer must confirm. The earlier the confirmation, the less rework later.
Note that this workflow applies to L2 and L3. L1 is completed on existing silicon, with no new mask set and no tapeout, so it skips the "Tapeout" and "First-Silicon Validation" nodes, moving from specification confirmation directly to engineering samples and measurement, which compresses the overall cycle to 4–8 weeks.
When to Customize and When Not To
Customization isn't a cure-all; it has a clear economic threshold. Pushing the wrong project into customization wastes the customer's money and burdens us. So this section spells out the reasons to walk away as well.
These Situations Justify Customization
- Annual volume can amortize NRE, masks, and tooling, and the total cost of customization is lower than that of "standard part + external parts + PCB changes."
- The pain point is specific and quantifiable: not "it feels expensive," but "four extra external components, too much board area, recertification required."
- A Pin-to-Pin replacement for the current part is needed to avoid PCB changes and recertification.
- There is a clear certification target that standard parts can't cover (such as a specific automotive AEC-Q100 grade or industrial wide temperature).
- The current part carries EOL or supply-interruption risk, and a controllable long-term supply plan is needed.
In These Situations, Hold Off for Now
- Requirements aren't frozen: while the specification keeps changing, any custom plan will be torn down and restarted.
- Annual volume is small, NRE and tooling can't be spread out, and the per-unit cost ends up far above a standard part.
- The only request is "a bit cheaper," with no real technical mismatch. This is better served by locking the part number and negotiating volume pricing.
- The project cycle is shorter than the target level's reference cycle, even L1's 4–8 weeks. Customization can't make the production date, so use a standard part for now and save customization for the next revision.
A common misjudgment is treating customization as a cost-reduction tool.
But customization has an economic threshold: if annual purchases can't cover amortization of NRE, masks, and tooling, it costs more than simply buying a standard part. Manufacturers in the industry have also publicly cautioned that for customers below a certain annual purchase volume, switching parts to cut costs doesn't pay off; it makes more sense to lock in the part number and watch the change-notice period.
Customization is no different: it solves "does it fit?", not "is it expensive?" To cut cost, talk volume and tiered pricing first. To resolve a mismatch, talk customization.
Five Pitfalls to Avoid in Customization
These five come from where real projects most often go wrong. Writing them into the plan up front saves many rounds of dispute later.
Pitfall 1Acceptance Criteria Must State Test Conditions
An accuracy figure without conditions can't be judged pass or fail at delivery. This is the most common source of dispute in custom projects.
Every electrical characteristic in the specification draft should carry conditions, a test method, and decision limits. "Typical" and "approximately" can't serve as a basis for acceptance.
Pitfall 2Pin-to-Pin Comes Before "Better Parameters"
The hidden cost of a PCB change (layout redraw, mechanical checks, recertification, production-tooling adjustments) is usually far greater than a small parameter gain. So the first question in the package and pin plan is always "can pin compatibility be achieved within the existing package family?"
If new tooling is truly needed, its investment and lead time belong in the commercial plan, not raised when design is nearly complete.
Pitfall 3Write Responsibility Boundaries into the Contract Before Work Starts
Two things must be settled up front: if first silicon misses the target (L2/L3), how many free respin rounds apply and how further rounds are charged; and if the customer changes the specification midway, how the schedule extends and how incurred costs are shared.
Leave these two out and each side will tell its own story later. Write them down, and both sides have an easier time.
Pitfall 4PCN / EOL / LTB: Don't Leave These Three Supply Terms at "Separate Notice"
Notice periods for Product Change Notification (PCN) and End-of-Life (EOL), along with the Last-Time-Buy (LTB) window, should all be written into the contract in days or months. Custom parts are inherently harder to replace than standard parts, and once one is discontinued, the customer has no second route. That makes the certainty of these three terms more important than price.
Pitfall 5A Specification Change Means a Schedule Extension
Freezing is a commitment; rolling back is a restart. Every parameter adjustment after freeze requires a fresh review of its knock-on effects on design, tapeout, and validation; a large change amounts to going back to the start and rescheduling. The sooner the customer thinks requirements through, the more time the project saves.
This is also why the "six questions" in Section 04 come before the quotation: moving uncertainty forward is the most cost-effective engineering practice there is.
Conclusion
Selection means finding the closest match among the existing options; customization means changing the option into the one you need.
The first tests your eye; the second tests whether both sides can state the specification, responsibilities, and timing clearly. What STST does is break "customization," which sounds so heavy, into three levels you can choose step by step: one parameter step off, take L1; the package doesn't fit, take L2; the system truly needs a new chip, then take L3.
From settling for an off-the-shelf part to a precise fit for your design. What lies between isn't a line saying "customization supported," but a plan that gets six requirements answered, fixes the acceptance criteria, and draws the responsibility boundaries.
References
- TrendForce: 2026 global power management IC market size and growth forecast.
- China Commercial Industry Research Institute: China analog chip market size and 2026 forecast.
- 36Kr: Stretched analog and power device lead times and manufacturer price adjustments.
- Sohu: Full 8-inch foundry capacity and rising quotes.
- AnySilicon: Lead time and process reference for custom PMIC design services.
- East Money: Custom analog foundry work and the "co-optimization" model.
- Elecfans: Economic thresholds for cost-driven part changes and part-number management advice.
FAQ
Does L1 really carry no NRE?
L1 adjusts parameter options and screening requirements on existing silicon, with no new mask set or design investment, so no NRE is charged. The customer bears packaging and test-related costs. The exact scope follows the feasibility assessment: if the review finds layout changes are needed, it is no longer L1.
Can the customization lead time be shortened?
Lead time is determined by process and validation steps; it isn't purely a management problem. L1's 4–8 weeks, L2's 8–16 weeks, and L3's 6–12 months are reference ranges under normal production scheduling and validation. Expediting needs a separate review and usually comes at the price of added investment or adjusted validation items. What can be compressed is usually queue time, not the silicon or the reliability tests themselves.
What is the minimum order quantity (MOQ)?
MOQ is tied to the level. L1 is usually set by packaging and test batches; MOQ for L2 and L3 relates to mask and tooling investment. The actual figure is worked out per project at the commercial-plan stage. We don't give a "universal number" in advance because it would be inaccurate for every project.
Can Pin-to-Pin compatibility be achieved?
This is the first question at the package and pin plan stage. STST prioritizes pin compatibility within existing package families. If new tooling is truly needed, its investment and lead time are included in the commercial plan so the customer knows the cost before work begins.
Can your measured data be trusted?
Measured reports are based on STST's own EVBs and test conditions, and state the test board, test points, instruments, and temperature conditions so the customer can reproduce them under the same conditions. We don't substitute datasheet typical values, competitor data, or simulation results for measurements. That is the fundamental difference between a custom service and "selection consulting."
Who owns the IP after customization?
There are two layers. STST's existing process platforms and background IP remain with STST. Design results newly created for the project are owned or licensed as negotiated according to the investment structure: an L3 project where the customer fully funds NRE and masks can agree on an exclusive license or joint ownership; L1 and L2 projects based on existing platforms usually give the customer a license within an agreed scope. Specific terms go into the contract, with no vague "default owner."
Can long-term supply be guaranteed after customization?
Supply assurance rests on three mechanisms: capacity commitments, advance notice periods for PCN and EOL, and an LTB window. All three are written into the commercial plan rather than left as verbal promises.