SiC Hiring in China: The Eight-Inch Experience Gap

Power Semiconductor Talent Research · Eight-Inch Ramp Watch

Seven eight-inch silicon carbide lines are ramping inside the same eighteen months, and the engineers to staff them do not exist yet. This report reads the job postings, the pay bands and the six sourcing routes, and it is written for the employer who has to file a requisition in China before the line is ready.

Published by SunTzu China | Data as of 23 October 2026

Source tiers: US Department of Commerce notice and financial media reporting / listed-company interim reports and results briefings / employer postings for live roles / broker and trade-press research on capacity and yield

Coverage: silicon carbide substrate growth, epitaxy, device process, process integration and module packaging

The China plant head of a European power-semiconductor group spent last week rewriting a requisition. The role is process integration on an eight-inch silicon carbide line, and the draft he inherited asks for eight years in SiC and five years on 200mm. Both figures were copied from the equivalent role in Germany, where they describe somebody who exists. In China they describe nobody, and pay is not the reason.

On 22 October the US Department of Commerce proposed $389m across four recipients: up to $175m for Helion, $85m for Commonwealth Fusion Systems, $73m for Vertical Semiconductor and $56m for Sibelco North America, covering fusion magnet drivers, radiation-tolerant electronics, vertical gallium nitride and ultra-high-purity quartz. Each is a letter of intent, with the government taking a minority stake without control or voting rights, under a $10bn research programme approved in 2022.

Read that list for what it buys. Fusion magnet drivers, radiation-tolerant electronics and quartz purification have no scaled line and no headcount inside China today. The Chinese engineers who work in silicon carbide sit in automotive traction inverters, solar and storage converters and charging piles. The overlap with those four awards is close to nil. A China plant that raises pay because of this announcement will spend the money on people who are not the ones leaving.

The pressure comes from the lines that started in the same twelve months. SMEC's Shaoxing line is due to reach 5,000 to 7,000 wafers a month by year-end against a target of 15,000. Changfei Advanced in Wuhan is doing ramp and qualification at once. Silan Jihong plans 35,000 a month in Xiamen, with full output in 2029. Sanan ships eight-inch SiC wafers from Changsha in small volume at 1,000 a month, with twelve-inch at customer sampling, the Sanan and STMicroelectronics venture is in risk production, and CRRC Times Electric is ramping.

Seven lines, one cohort of people, and a wafer size that almost nobody in the country has worked on for longer than three years.

Table 1 basis: nameplate figures come from company announcements and results briefings; blanks are left blank rather than estimated. The "monthly equivalent" figure is a broker and trade-press estimate, the methodologies of which are not aligned, so it is used for scale only and never as a denominator.

The price line does not follow seniority. It follows how many process steps a person can see at once. A role that owns one step in crystal growth posts at RMB 12,000 to 16,000 a month; a role that owns one step at the device front end posts at RMB 20,000 to 40,000; a role that holds substrate defects, epi uniformity and the device process window in one view posts at RMB 25,000 to 50,000 on fourteen months. Years in SiC move the number, and they move it one tier less than process coverage does.

The two ends of the chain price in opposite directions. Substrate capacity is tight and prices are steady, with SICC's eight-inch conductive substrate now above half of revenue. The device end is squeezed: StarPower reported first-half revenue of RMB 1.928bn, down 0.41%, net profit of RMB 65m, down 76.40%, and operating cost up 12.58%, which the interim report attributes to higher raw material cost and to earlier price cuts.

Four figures cover most of what will reach a hiring desk this cycle. A master's graduate entering crystal growth is posted at RMB 12,000 to 16,000. Five years of eight-inch crystal growth is RMB 12,000 to 24,000. Eight-inch device process is RMB 20,000 to 40,000 on fourteen months. Process integration is RMB 25,000 to 50,000 on fourteen months. The remaining spread comes from city and from whether the candidate has carried a project through an automotive qualification.

Nature of data: pay figures are employer-stated monthly ranges taken from live postings, shown on thirteen or fourteen months where the employer says so and on twelve where it does not. Capacity and yield figures come from company announcements, results briefings and trade-press research; where a figure is an industry estimate it is labelled. Sample sizes within a single city and job title are too small for distributional statistics, so no percentiles are given anywhere in this report. Currency is renminbi throughout and is not converted.

I. Why the specification describes nobody

The first thing that breaks on a China requisition is the job title. Crystal growth appears as crystal growth, substrate process, PVT process or growth engineer. Epitaxy appears as Epi, epitaxial growth or epi process development. The device front end appears as process engineer, process engineering, etch process or implant process. Integration appears as PIE, process integration or TD process. Module work appears as packaging process, module engineering or power module. A search built on one title reaches perhaps half the people who can do the job.

Search by skill word instead. For growth: PVT, seed crystal, defect density, micropipe density, growth rate. For epitaxy: uniformity, doping concentration, surface roughness, high-temperature anneal. For the device front end: implant, activation anneal, gate oxide, trench etch. For qualification: AEC-Q101, IATF16949, MSA, SPC. A dozen words of this kind reach further than a title list, and they bypass the grade names each company invents for itself.

Four lines in the requisition decide what quality of candidate replies. Which generation of line, and whether the candidate saw it ramping or at full output. Substrate defect density and its effect on device yield, written as how far the candidate moved it. The depth of automotive qualification work, where an AEC-Q101 submission reads a tier above being familiar with the standard. Position inside IATF16949, as auditor or as the party audited. Written vaguely, these four return a stack of six-inch lateral moves.

A crystal growth engineer at Sanan's Changsha site is posted at RMB 12,000 to 16,000 a month for a master's in materials or semiconductor physics, with duties that read as holding the line process, writing the station SOP, handling excursions and running process improvement. The same site posts five years of SiC crystal growth at RMB 12,000 to 24,000 and adds one line to the requirements: eight-inch experience preferred. Read together, the eight-inch premium at that tier is roughly thirty per cent. It is smaller than most groups assume when they set a band.

The device front end sits a tier higher, because of proximity to the customer's qualification window rather than difficulty. A device fab in Chongqing posts SiC etch development, process integration, eight-inch process and metrology at RMB 25,000 to 50,000 and RMB 20,000 to 40,000, all on fourteen months and all asking a bachelor's with five to ten years. The same degree and years buy RMB 12,000 to 24,000 in crystal growth. The gate is identical; the distance from a carmaker's approval is not.

A process integration engineer at the top of that band is paid for three judgements at once: whether substrate defect density can support the yield the line promises, how far epi uniformity varies within and between wafers, and where the device process window can be set without giving up reliability. Get one wrong and yield stops. Very few people in China hold all three, because the eight-inch lines themselves are at most two years old. The SiC process engineer salary band at the top is set by that scarcity, not by a grade.

The equipment side prices differently. Nanjing Jingsheng Equipment posts a SiC epitaxy process engineer at RMB 20,000 to 40,000 for five years of epitaxy or two of MOCVD, with on-site commissioning at customer fabs. That is close to device-fab pay, and the premium is for making a process work on somebody else's line. To a device fab these engineers are short-term capacity; to an equipment maker they are part of the product, which is why they are rarely available.

Table 2 basis: monthly ranges are as published by employers and are not estimates of actual pay. Fourteen-month postings are shown as stated; the rest are annualised on twelve. Within one city and one title the public sample runs to one or three postings, so the table supports scale and not distribution. Module packaging is left blank because no verifiable comparable posting was found, and no adjacent role is substituted for it.

SiC module packaging is the one segment this report cannot price. The work is to hold die, solder, substrate and terminal inside a single thermal-cycle life, and silicon carbide's high-temperature behaviour makes the solder and substrate choices harsher than they are for silicon. The people who can do it come largely from IGBT module lines and from carmaker inverter teams. Employers in this segment report paying above device process rates, and the figure is not published anywhere verifiable. A plant that budgets module packaging from the device process band will be short.

Automotive qualification is the other gate, and it is not one document. AEC-Q101 covers device-level reliability, IATF16949 covers the system, and a customer audit runs three to six months on top of both, during which somebody has to sit with the auditors and answer process-capability questions station by station. That person, in most Chinese fabs, is a part-time assignment for an engineer who already has a full workload, and it is usually the variable that decides whether qualification lands on schedule.

II. Where the people come from when the pool is three years old

Chinese six-inch silicon carbide lines only scaled from around 2019. The first eight-inch lines commissioned in 2025. Lines that have run two full years can be counted on one hand. There is no candidate in the country with eight years on eight-inch, and the CVs that claim five years and above in SiC are, with rare exceptions, six-inch CVs. Any specification built on years of eight-inch experience is asking for a person the market has not produced.

What exists instead is six channels, each with its own price and its own speed.

Internal conversion is the fastest and the cheapest. Equipment physics, the process parameter system and defect classification are the same set at six inches and at eight; what has to be added is thermal field uniformity and stress control at the larger diameter. A converted engineer is productive on the new line in months, while an external hire takes quarters to arrive and more to be trusted with a process window. Plants that convert first and search second usually staff a ramp faster than plants that do it the other way round.

Equipment vendors are the second route and they are not for sale in the usual sense. Application engineers at furnace, epi reactor and high-temperature implanter makers spend their weeks commissioning processes on customer lines, so they have seen more eight-inch problems than most fab staff. Their number is small, their price is high, and they mostly do not want to leave. Project-based cooperation plus a long consulting arrangement is the structure that works.

Returnees are the third route and the one where a foreign employer holds an advantage. Wolfspeed's debt restructuring released teams, and the people among them who have run an eight-inch ramp are scarce inside China. The obstacle is structure rather than total: an overseas package is base plus equity plus relocation, and most Chinese employers offer base plus annual bonus. A group that can offer the structure it offers in Europe starts this conversation a step ahead.

Silicon fabs supply part of the device front end. Metrology, implant and high-temperature anneal transfer directly, because the tools and data systems are close. Crystal growth and epitaxy do not: thermal field design and vapour-phase growth have to be rebuilt. This route fills device process and cannot fill substrate, and treating it as a general source is how a substrate requisition ends up with a hundred irrelevant CVs.

Research institutes are the slowest route and the main source for growth and epitaxy. The Chinese Academy of Sciences groups working on silicon carbide, the wide-bandgap programmes at Shandong University and Xidian University, and the device line at the 55th Institute all produce people who understand the physics. Their weakness is production tempo: years of small batches and paper-style validation, followed by six months in a fab learning the language of yield and cost. Budget that half year or the hire will look like a mistake at month three.

Campus supply is the sixth route and covers only part of the org chart. Growth and epitaxy take master's graduates in physics, materials and microelectronics who can hold a shift after two supervised rounds. Device process and integration cannot be built from campus, because the job is judgement under incomplete information. The calendar is fixed: the final autumn semester, September and October, locks a July start, and the three-party agreement is signed then. A plant that decides in spring waits a full year for the same cohort.

The training arithmetic is what makes all of this urgent. A six-inch process engineer moving to eight-inch needs one full automotive qualification cycle plus one yield iteration before they can judge a process window alone, which is eighteen to twenty-four months. A line takes less than that to go from commissioning to full output. A plant that waits until full output to start converting will be a generation short at exactly the moment it needs the yield.

Qualification and ramp want different people, and one requisition cannot cover both. Qualification wants APQP, PPAP, customer audits and 8D reports, with the work sitting in documents and in front of the customer. Ramp wants people who will cover shifts at the tool, isolate a fault quickly and lift utilisation from seventy to ninety per cent. The experience does not transfer between the two seats. Split them, or the posting attracts neither.

Conversion also fails on grade more often than on money. The skill moves across the wafer size; the job ladder often does not, and the person is slotted at the entry grade of the new role. Take-home falls, the candidate declines, the plan restarts. Keeping the original grade and putting the increment into a project bonus tied to yield costs less than an external hire at the top of the band.

The geography follows industrial belts rather than provinces. Substrate sits in Jinan, Shanghai Lingang and Changsha; epitaxy in Xiamen and Suzhou; devices in Shaoxing, Zhuzhou, Xiamen, Chongqing and Wuxi; equipment in Nanjing, Suzhou and Shenyang. Shaoxing to Zhuzhou is a cross-province relocation and Shaoxing to Hangzhou is a longer commute, which is why a relocation package drawn on provincial lines will be mispriced.

Power semiconductor hiring this year has shifted toward commissioning whole process crews rather than single roles. Half the projects reaching a search firm are scoped by shift or function group, because a new line needs a full crew and one missing link stops the output.

Four kinds of employer bid for the same people with four different pitches. Device IDMs sell the whole-chain view and stable orders. Substrate makers sell material scarcity and pre-IPO equity. Equipment makers sell field allowances and cross-customer exposure. Carmaker and Tier 1 inverter teams sell proximity to the end product. A competitor map built without those four columns produces one pitch against four audiences.

The SiC talent pool is small enough that conference lists beat databases. The process engineers in this field appear at the same handful of events year after year. Three years of exhibitor and speaker lists is a talent map of most of the market. A power semiconductor recruiter is better judged on whether it can produce that list than on how many CVs sit in its database, and the question is worth asking before a search is awarded.

III. When to commit, and what the band is really competing against

The substrate end is tight on capacity, steady on price and short on yield. SICC's Lingang and Jinan sites carry heavy early investment and the depreciation from it, utilisation is still climbing, and at the September briefing the company said capacity would be tighter in the second half. The headcount logic there is direct: one more good wafer is one more unit of revenue against fixed depreciation.

The yield gap explains the urgency. Global leaders run above 85 per cent on eight-inch substrate, the Chinese top tier sits at 75 to 80, and SICC is put near 70 by trade-press and broker estimates. The cause is thermal field design, defect control and process stability rather than tool purchase. Every point is bought with repeated trials, and trials need engineers.

The price increases have not passed through evenly. Sanan Optoelectronics told investors it had raised some power electronics chip prices against higher precious metal and substrate cost, while SICC said substrate prices have been broadly stable since January. At the device end cost is outrunning revenue, so the upstream increase has reached the cost table while carmaker pricing holds the selling price down. One end bids for people and the other holds headcount.

Eight-inch economics only open up once yield arrives. The working rule in the industry is that equivalent cost per die has to fall below 0.7 times the six-inch figure before the wafer-size arithmetic pays. At seventy per cent yield that multiple is not reached, and depreciation eats the gain. That single ratio ties yield to cost and puts the people who can move yield into the middle of the budget conversation, which is not where a process engineering function usually sits.

Sanan's own language on the device side is worth reading closely. The company said first-half 2026 silicon carbide utilisation was not high and that product mix improvement is incomplete, with the priority on raising utilisation and on the profitability of the integrated circuit business. Translated into hiring, the line is not full, so lifting utilisation beats adding capacity, and headcount stays flat while the skill mix changes: six-inch operators and process staff converting to eight-inch, which is internal upskilling and not a recruitment line item.

The technical step is already moving. Sanan's twelve-inch silicon carbide is in validation at a lead customer while the eight-inch line has only just settled. The volume is tiny and confined to pre-production lines, and it still decides what the talent structure looks like in three to five years. A group building a five-year China plan should put a marker on that seat now, even with nothing to hire for this year.

Put the two ends on one page and the price elasticity of this cycle sits at the substrate end and in integration. Substrate pays a premium for a point of yield because depreciation presses on it. Integration prices off a supply gap held by very few people. Routine device process roles face margin compression, and their negotiating room is narrowing rather than widening.

The timing window is specific. Most of the seven lines complete their ramp between 2026 and 2027, and Silan Jihong's Xiamen phase one does not reach full output until 2029. Before 2027 the supply of eight-inch experience grows more slowly than demand and the band moves up. After 2027 the first cohort that ran a complete ramp inside China appears on the market, and the price for that same profile flattens. Locking people now costs measurably less than locking the same people in two years, and the difference is larger than any annual merit budget.

This is where a group process usually breaks. A band written for a process integration engineer in Europe will look, on paper, like it overlaps the top of the Chinese range. The overlap sits in the number and not in the person: at the top of the Chinese range the candidate has three years on eight-inch and no completed qualification cycle, while the European holder of that band has ten. Two adjustments follow. Hold the band and move the search down a grade into people who can be trained, budgeting the training openly. Or open the band and write the requisition around line generation and qualification depth instead of years. Holding the band while asking for ten years leaves the vacancy open two quarters and fills it at the top of the band anyway.

Four numbers tell you which way the cycle is going, and each of them comes from a company's own disclosure rather than from an estimate.

Table 3 basis: all four indicators are obtainable from public disclosure and are used to read the direction of the band, not to estimate a talent shortfall. The yield and monthly-equivalent figures quoted in this report are industry estimates and may differ from company disclosure; where the two differ, company disclosure governs.

The most direct limitation is sample size. The pay figures come from roles currently posted, and within a single city and title there are often one to three. That supports a range and a scale, and not a median. Actual pay also includes project bonus, equity and field allowances that are not published.

The blanks in the tables are real gaps rather than rounded estimates. SiC module packaging produced no verifiable comparable posting, and no adjacent role was used to stand in for it. That makes the module conclusion thinner than the others, and a plant that budgets from it should carry the uncertainty explicitly rather than average it away.

The yield figures carry a second-order problem. The 85, 75 to 80 and roughly 70 per cent levels come from trade-press and broker reporting, and none of those sources states whether automotive and industrial grades are on the same basis. They read direction. They do not support a computed gap.

The $389m is still a letter of intent. All four arrangements await final funding agreements, the amounts are ceilings rather than disbursements, and the government's participation is a minority equity position whose terms are not settled. The judgement here concerns where the money points, not how much of it will be spent. If a final agreement changes the technical direction, the opening reading of this report has to be revised.

Headcount figures from different companies are not on the same date. CRRC Times Electric reported 10,746 employees, up 7.35 per cent, with 4,365 research staff at 40.71 per cent of the total and postgraduates at 52.58 per cent of research staff, in its Hong Kong interim report. StarPower's 782 research staff at 23.69 per cent is an end-2025 figure and does not match the mid-2026 headcount period. The two are used only for internal comparison within each company and are not divided against each other.

What this means for a foreign principal is one ordering decision. The people worth committing to now are those who can hold two or more process steps in a single view, because that seat is where the price moves. Engineers who have simply worked in silicon carbide come after. The first group numbers under a hundred nationally and sets its own price; the second is a broad pool at a transparent rate. Decide which step is missing, then whether to convert or to hire, then the price. Reversed, a plant pays the right price for the wrong person and finds out at the first yield review.

SunTzu China

Executive search for advanced manufacturing, semiconductors and new energy. All figures cited come from public sources and do not constitute investment advice. Pay bands are employer-stated ranges, and any individual offer rests on the employer's own assessment.