Burn-in & reliability (process step)

Step 55 of 56 in the semiconductor flow (FinalTest). Prev: Final test (ATE) (process step) · Next: Marking, tape & reel, final inspection (process step)

What this step does

Burn-in stresses parts at elevated temperature and voltage (typically 125-150C, sometimes higher current/voltage) to force “infant-mortality” failures, the weak devices that would otherwise die in the first weeks of field use, out before shipment. Reliability testing extends this to lifetime characterisation. It is the screen that turns a part that passes electrical test into a part trusted in a car, a satellite, or an AI server rack. Sub-processes inside this step:

• Package-level burn-in (PLBI): packaged parts loaded into burn-in boards inside ovens/chambers, powered and heated for hours.
• Wafer-level burn-in (WLBI): stress applied to the whole wafer before dicing, the route to known-good-die.
• High-temperature operating life (HTOL) and other reliability quals (HTRB, temperature cycling, HAST) for qualification lots.
• Dynamic / monitored burn-in, where parts run real patterns and are watched for failure, not just baked.

Where it sits and why it matters

This sits after Final test (ATE) (process step) and before Marking, tape & reel, final inspection (process step). For commodity logic it is often skipped or sampled. It becomes mandatory the moment a die is going into something expensive or unforgiving: SiC and GaN power devices for EVs, automotive-grade parts, and now AI accelerators and HBM stacks. The economic driver is brutal arithmetic: in a 2.5D/3D package with eight HBM stacks plus a compute die, one latent failure scraps a module worth thousands. That pushes stress-screening earlier in the flow (“shift left”) and onto the wafer, which is the Known Good Die Test Wedge.

Equipment market

Burn-in systems (ovens, drivers, WLBI tools) are a small, fragmented niche inside test. Independent estimates put the burn-in test system market at roughly $0.8bn in 2025 growing ~8-10% CAGR [1][2]; the broader “wafer-level burn-in & reliability” bucket (including sockets, contactors, probe cards) is sized far larger at ~$2.3bn in 2025, 15% CAGR, but that figure double-counts consumables and contactors [3]. Treat the pure-systems pool ($0.8-1bn) as the honest number.

Concentration: “competitive and fragmented… no single company holding a dominant share” [4]. WLBI is the exception, where Aehr is effectively the scaled pure-play.

Materials & consumables

The recurring spend lives below the system: burn-in boards (custom PCBs that fail and get replaced), and the sockets/contactors that touch each part. The burn-in & test socket market is ~$1.5bn+ and growing ~6-7% CAGR, with Yamaichi, Cohu, and Smiths Interconnect together ~40% of it; Enplas, Aries, 3M follow [6][7]. WLBI adds wafer contactors and full-wafer contact assemblies. These are genuinely recurring (sockets wear, boards are device-specific), so the consumable layer compounds with unit volume in a way the capital-tool layer does not.

Volumes, revenue, profitability

There is no clean “units” metric; volume tracks the share of devices that must be screened (automotive, power, AI/HBM), not total wafer starts. Pool: systems ~$0.8-1bn, sockets/boards ~$1.5bn, so the whole step is a low-single-digit-billion revenue pool inside a ~$8-9bn test-equipment market. Margins: the value-capturing layer earns well. Aehr in its 2024 SiC peak posted non-GAAP net income of $35.8m on $66.2m revenue [8]; the 2025 diversification year softened to $59.0m revenue, GAAP net loss $(3.9)m, non-GAAP net income $4.6m as it absorbed lower-margin Incal product [9]. So the economics are gross-margin-rich (50%+ typical for the leaders) but operationally lumpy and order-driven at this scale. The socket/board vendors earn steadier consumable margins.

Competitive landscape & value capture

The moat is qualification, not raw technology: getting a burn-in tool and socket recipe designed-into a customer’s automotive or AI flow is sticky and slow to displace, which is why incumbents persist and why Aehr’s AI-processor design-wins (lead hyperscaler ASIC, silicon-photonics WLBI follow-on orders in 2026 [5][10]) matter more than headline TAM. Value accrues to (a) the WLBI tool owner where shift-left is real, and (b) the consumable socket/contactor specialists who earn on every device. The vulnerability is that burn-in is a cost customers route around when yields are mature, so demand is concentrated in the hard cases.

Market drivers, constraints & trends

Net read: the served pool grows steady-to-accelerating, but the acceleration is concentrated in two pockets (AI optical I/O and HBM/KGD), while the old SiC growth engine is in a cyclical trough through 2027.

• Drivers

• AI/datacentre is now the swing factor. Aehr booked $37.2m in Q3 FY2026 at a book-to-bill above 3.5x, guided H2 bookings to the high end of $60-80m, and ran one AI customer at 42.1% of Q3 revenue against an effective backlog of $50.9m [11][12][18]. The growth is moving from SiC to AI processors, HBM and silicon photonics.

• Silicon photonics / optical I/O is the newest WLBI line: Aehr’s H2-2026 follow-on orders ship FOX-XP systems rated up to 3,500W per wafer (the June 2026 order tests nine wafers in parallel) for data-centre optical interconnects, a segment that barely existed two years ago [13][19].

• HBM volume pulls reliability spend: HBM grew ~130% in 2025 and is forecast ~70% in 2026, and HBM4’s 16-high stacks push a known-good-die requirement that mandates wafer-level burn-in before dicing [14].

• One latent die in a $30k+ HBM4 wafer or a multi-thousand-dollar 2.5D module makes shift-left screening economically forced, not optional [15].

• Constraints

• Burn-in equipment overcapacity is capping the aggregate: independent reads put burn-in system growth at only ~3% CAGR despite the AI pull, because the tool base over-built into the 2019-24 SiC cycle [16].

• SiC, the prior demand engine, is in a downturn to 2027-2028 with upstream utilisation near 50% and the EV ramp delayed [17].

• Burn-in is a cost customers route around as yields mature, so demand stays concentrated in hard cases (auto, power, AI/HBM) rather than broadening (see §Competitive landscape above).

• Customer concentration: the listed pure-play’s AI exposure rests on one or two hyperscaler-ASIC accounts, so a single design slip swings the number.

• Trends & inflections to watch

• Thermal redline: AI processors now need 2.5-3.5kW per-wafer dissipation, forcing liquid-cooled chucks and kiloamp probe cards. Tripwire: a second WLBI vendor shipping a >3kW liquid-cooled production tool would confirm the optical/AI line is multi-sourced and structural [13][14].

• HBM4 stays on microbumps; hybrid bonding (sub-10um) slips to HBM4E ~2027. Tripwire: a foundry/OSAT qualifying hybrid-bonded HBM in volume re-rates the KGD/WLBI requirement upward [14].

• Burn-in-board pull as a leading indicator: Trio-Tech booked a $5.3m AI-GPU burn-in-board order in March 2026. Tripwire: repeat eight-figure board/socket orders tied to named AI platforms confirm the consumable layer is compounding ahead of tools [15].

Connections

• Wafer-Level Test & Burn-In (WLTBI) — the shift-left technique central to this step
• Known Good Die Test Wedge — the thesis: KGD/WLBI goes mandatory for 2.5D/3D and AI dies by 2028
• Final test (ATE) (process step) — upstream electrical test
• Marking, tape & reel, final inspection (process step) — downstream finishing

Sources

1. SNS Insider, Burn-In Test System for Semiconductor Market — https://www.globenewswire.com/news-release/2026/01/13/3217840/0/en/Burn-In-Test-System-for-Semiconductor-Market-Size-to-Grow-1530-25-Million-by-2033-SNS-Insider.html
2. Market Report Analytics, Burn-in Test Equipment ~$756m 2025, ~9.9% CAGR — https://www.marketreportanalytics.com/reports/burn-in-test-equipment-for-semiconductor-394837
3. Virtue Market Research, Wafer-Level Burn-In & Reliability Testing (~$2.26bn 2025, 14.8% CAGR) — https://virtuemarketresearch.com/report/wafer-level-burn-in-reliabilit-testing-market
4. SemiconductorX / market vendor lists (Advantest, Aehr, Chroma, Teradyne, Incal, Espec) — https://www.24marketreports.com/semiconductor-and-electronics/global-burnin-test-system-for-semiconductor-forecast-market
5. Aehr, momentum in package-level burn-in alongside WLBI (Incal acquisition, capacity) — https://www.aehr.com/2025/11/aehr-seeing-momentum-in-package-level-burn-in-alongside-expanding-wafer-level-burn-in-demand/
6. Yole Group, Burn-in and Test Sockets Market Monitor — https://www.yolegroup.com/product/quarterly-monitor/semiconductor-test-consumables---burn-in-and-test-sockets-market-monitor/
7. Global Growth Insights, Test & Burn-in Sockets (Yamaichi/Cohu/Smiths ~40%) — https://www.globalgrowthinsights.com/market-reports/semiconductor-test-and-burn-in-sockets-market-115895
8. Aehr Test Systems FY2024 results — https://www.sec.gov/Archives/edgar/data/0001040470/000165495424009008/aehr_ex991.htm
9. Aehr Test Systems FY2025 full-year results ($59.0m rev, GAAP net loss $(3.9)m) — https://www.aehr.com/2025/07/aehr-test-systems-reports-fiscal-2025-fourth-quarter-and-full-year-financial-results-expands-total-addressable-market-and-diversifies-customer-base/
10. Aehr, follow-on WLBI order from silicon-photonics customer (Jun 2026) — https://www.aehr.com/2026/06/aehr-receives-follow-on-order-from-major-silicon-photonics-customer-for-fully-automated-wafer-level-burn-in-system-for-hyperscale-data-center-optical-interconnect/
11. Seeking Alpha, Aehr targets high side of $45-50m FY2026, effective backlog $50.9m — https://seekingalpha.com/news/4573152-aehr-targets-high-side-of-45m-50m-fy2026-revenue-as-effective-backlog-reaches-50_9m
12. Aehr Q3 FY2026 8-K (Customer A 42.1% of Q3 revenue, $37.2m bookings) — https://www.sec.gov/Archives/edgar/data/1040470/000165495426003310/aehr_ex991.htm
13. Aehr, follow-on WLBI order for AI optical I/O / silicon photonics, FOX-XP up to 3,500W/wafer, ships H2-2026 (Mar 2026) — https://www.aehr.com/2026/03/aehr-receives-follow-on-order-for-fully-automated-wafer-level-burn-in-systems-powering-ai-optical-i-o-and-data-center-interconnects/
14. Virtue Market Research, WLBI & reliability ~$2.26bn 2025 → $4.51bn 2030 (14.8% CAGR); HBM ~130% 2025 / ~70% 2026, HBM4 KGD mandates WLBI, 2.5kW+/wafer liquid-cooled chambers — https://virtuemarketresearch.com/report/wafer-level-burn-in-reliabilit-testing-market
15. Temple8 Capital, the burn-in tax: $30k wafers, shift-left economics, Trio-Tech $5.3m AI-GPU burn-in-board order (Mar 2026) — https://temple8capital.substack.com/p/semiconductor-burn-in-testing-ai-hbm4-reliability-stocks
16. IDTechEx, power electronics market — burn-in system overcapacity offsets growth, ~3% CAGR — https://www.idtechex.com/en/research-report/power-electronics-market/1152
17. Yole Group, Power SiC overcapacity downturn to 2027-2028, utilisation ~50%, device market to ~$10bn by 2030 — https://www.semiconductor-today.com/news_items/2025/dec/yole-181225.shtml
18. Aehr Test Systems, Q3 FY2026 over $37.2m bookings, book-to-bill >3.5x, H2 bookings to high end of $60-80m, lead hyperscaler PLBI + CoWoS/HBM WLBI interest — https://www.aehr.com/2026/04/aehr-test-systems-reports-over-37-million-in-quarterly-bookings-driven-by-strong-ai-and-data-center-infrastructure-demand/
19. Aehr Test Systems, follow-on production order for FOX-XP testing nine wafers in parallel from a data-centre optical-transceiver silicon-photonics customer (Jun 2026) — https://www.aehr.com/2026/06/aehr-receives-follow-on-order-from-major-silicon-photonics-customer-for-fully-automated-wafer-level-burn-in-system-for-hyperscale-data-center-optical-interconnect/