Aehr expands Sonoma AI processor burn-in

Aehr expands Sonoma AI processor burn-in

Aehr has secured additional Sonoma orders for AI processor burn-in. The $6 million package supports a second processor programme ahead of a planned 2027 production ramp.


IN Brief:

  • Approximately $6 million of additional orders cover Sonoma systems, burn-in modules, sockets and production software.
  • The equipment will support the same hyperscale customer’s second AI processor programme at an OSAT facility in Taiwan.
  • Deliveries are scheduled across six months, with production of the second device expected to ramp during 2027.

Aehr Test Systems has received approximately $6 million of additional production orders for its Sonoma package burn-in platform, supporting a hyperscale customer’s second AI processor programme ahead of an expected manufacturing ramp in 2027.

The order covers additional Sonoma test and burn-in systems, turnkey burn-in modules, sockets designed for the processor package and production software enhancements. The equipment and consumables will be installed at one of the customer’s outsourced semiconductor assembly and test partners in Taiwan, with deliveries scheduled over the next six months.

This is the second processor from the same customer to use Sonoma for production burn-in. The first device is already in volume manufacture, while the newer processor has moved beyond the initial production award into additional capacity. Aehr says the customer expects the second device eventually to reach higher production volumes than the first.

Package burn-in applies electrical and thermal stress after the semiconductor has been assembled into its finished package. The process is intended to expose devices that are likely to fail early, including faults associated with silicon, interconnects or the completed package. By this stage of manufacture, substantial value has already been added through advanced packaging, memory integration and final assembly, making the cost of passing a weak device downstream particularly high.

Large AI processors make the thermal problem more complicated because a package can contain several functional regions with different power densities. Compute dies, memory and interfaces can dissipate heat differently while the device is operating. Treating the complete package as one uniform thermal load can therefore leave some regions above or below the stress condition intended by the test programme.

Aehr’s Sonoma architecture addresses that problem with independently controlled thermal zones across the package. The test system can manage those regions while applying electrical vectors and monitoring device functions, giving the production process more control over where heat is generated and removed. That approach is intended for processors whose power and package complexity make a single temperature condition increasingly difficult to maintain.

The burn-in module and socket are also part of the engineering system rather than incidental consumables. The socket has to carry the required current through reliable contacts while repeatedly loading and unloading expensive packages. The module has to distribute power, signals and thermal control around the device without introducing enough resistance or mechanical variation to distort the test result.

A new processor generation can therefore require different interface hardware even when the factory continues using the same Sonoma base platform. Changes in package dimensions, contact pitch, power requirements and thermal behaviour alter the equipment that sits between the tester and the semiconductor. The latest $6 million order includes those processor specific elements alongside additional systems.

Aehr also supplies wafer burn-in through its FOX platform. Wafer testing occurs earlier, before individual die have been packaged, and can remove weak silicon before money is spent on assembly. Package burn-in addresses the finished device after package interconnects and other integrated components have been added. Manufacturers can use both stages where the economics and reliability requirements justify screening at several points in the process.

The customer has already committed Sonoma capacity for the first AI processor and placed an initial production order for the second earlier this year. The new order therefore expands an established programme rather than starting a fresh evaluation. Aehr expects deliveries over six months as the customer prepares for increased production in calendar 2027.

No customer name, processor specification or exact factory capacity has been disclosed. Aehr has said the newer device is more demanding and is expected to reach larger production volumes, but the $6 million order should be read as additional test capacity rather than a direct measure of eventual processor shipments.

The manufacturing requirement is nevertheless concrete. A second AI processor family is moving through the same package burn-in architecture, and the new systems, modules and sockets will expand the Taiwanese OSAT installation before the planned production increase begins.


Stories for you


  • AVerMedia integrates JetPack 7.2 vision stack

    AVerMedia integrates JetPack 7.2 vision stack

    AVerMedia and Stereolabs aligned production vision software across Jetson platforms. JetPack 7.2 support covers ZED X cameras, carrier boards, drivers and reproducible software packages.


  • Aehr expands Sonoma AI processor burn-in

    Aehr expands Sonoma AI processor burn-in

    Aehr has secured additional Sonoma orders for AI processor burn-in. The $6 million package supports a second processor programme ahead of a planned 2027 production ramp.