Aehr secures $22m AI wafer burn-in order

Aehr Test Systems has received a $22 million follow-on production order for wafer-level burn-in equipment from its lead AI processor customer, extending full-wafer reliability screening before advanced processors enter increasingly expensive packaging flows.

The order covers multiple fully automated FOX-XP wafer-level burn-in systems, proprietary WaferPak full-wafer contactors, and integrated WaferPak Auto Aligners. Each FOX-XP system will be configured to test and burn in nine 300mm wafers simultaneously, with shipments planned over the next six months for installation at the customer’s high-volume manufacturing partner in Taiwan.

Aehr has not identified the processor supplier, describing it as a leading manufacturer of advanced AI training and inference devices. The more useful detail is where the screening takes place: defective or marginal processor die can be identified while they remain on the wafer, before singulation and before high-bandwidth memory, interposers, substrates, and other costly packaging elements are committed.

Burn-in applies controlled electrical and thermal stress to expose early-life or latent device failures that may not appear during conventional production test. The method is well established, but high-power AI processors make wafer-level implementation demanding because the equipment must deliver substantial current, control temperature, maintain reliable electrical contact across the wafer, and handle many devices in parallel.

Advanced packaging changes the economic calculation. A processor die that fails before assembly represents the value already invested in fabrication and test; the same failure discovered after HBM, substrate, interposer, and packaging operations have been added can write off a considerably more expensive assembly.

Additional screening is not free. Burn-in consumes equipment capacity, adds process time, and can lengthen the manufacturing cycle, so semiconductor suppliers have to weigh the cost of testing against the probability and downstream cost of failures that would otherwise escape into packaging.

Aehr’s latest order suggests its lead customer has decided that balance favours more capacity. This is a follow-on production purchase rather than an evaluation system, indicating that the FOX-XP process is already being used within the customer’s AI processor manufacturing flow.

The platform is built around full-wafer parallelism. WaferPak contactors provide the electrical interface to the wafer, while the automated alignment equipment positions it for repeatable high-volume operation. In the ordered configuration, each FOX-XP system handles nine 300mm wafers simultaneously.

Parallel operation addresses one of burn-in’s least convenient characteristics: useful stress takes time. Simply shortening the cycle risks reducing its ability to expose marginal devices, while running more wafers at once raises throughput without requiring the same reduction in test duration.

Automation becomes equally important once the process moves beyond engineering lots. Repeated wafer loading, contact alignment, thermal control, recipe execution, and unloading have to occur with minimal operator intervention if reliability screening is not to become another variable in the production line.

The destination of the equipment also reflects the outsourced nature of advanced semiconductor manufacturing. The systems will be installed at the customer’s high-volume manufacturing partner in Taiwan rather than at the processor company’s own facility, placing burn-in alongside the external fabrication, test, and packaging capacity used to assemble the final devices.

Aehr markets the FOX platform beyond AI processors, including silicon photonics, power semiconductors, memory, sensors, and microcontrollers. AI devices provide a particularly strong case for pre-packaging screening because individual packages can combine several high-value semiconductor technologies whose cost is lost if one unqualified processor die fails after assembly.

The $22 million order will therefore add capacity quickly compared with a new fab or packaging-line build, with deliveries spread over six months. Whether the approach becomes standard practice across a wider group of AI processors will depend on yield, field reliability, package value, and the throughput penalty associated with the additional test step.

Advanced packaging has increased the price of discovering a bad die late. Aehr’s latest order is evidence that at least one AI processor manufacturer is prepared to spend substantially more on finding those failures while the devices are still on the wafer.