Product1 publisher3 min readPublished
Onsemi's 1MW rack argument rests on a single circuit-breaker measurement
Onsemi says racks draw about 60kW today and will need 1MW by the end of the decade, and its Embedded Power Platform puts power dies inside the silicon wafer to fit that load into the same rack space. Samples go out this year, revenue in 2027.
The Product Desk · Product desk
What happened
- In an early EPP-based solid-state circuit-breaker design, the company says the result was about 50 percent smaller and 20 percent cooler than existing designs.
- Subaru is the only named partner so far, and Ramanathan said Onsemi is also working with an industrial company and an AI data center company that he could not name.
- Ramanathan said Onsemi is sampling customers this year and working towards revenue in the 2027 time frame, with qualification expected in Q1.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- constraint If the projection holds, a rack has to carry 16.7 times the power inside a volume that does not grow, so every millimetre of packaging overhead in the power path is space a converter cannot occupy.
- decision Anyone taking samples is deciding on a component comparison produced by the vendor, because no rack-level or third-party number for EPP exists yet.
- constraint The 200kW generation will be specified and bought before EPP is in volume, so those racks get designed around packaging that exists now.
- capability Putting FETs, drivers and controllers into one co-optimized package on standard 12-inch lines moves thermal design work from the board team to the package supplier.
A power engineer specifying a 200kW rack is working inside a volume that was set when racks drew 60kW [3]. Onsemi's one quantified EPP result is a circuit breaker. In an early solid-state circuit-breaker design, the company says the part came out about 50 percent smaller and 20 percent cooler than existing designs [8]. That saving is worth having in a space that cannot grow, though it covers one part in one early design and Onsemi measured it itself. The company did not publish a rack-level power density figure [3].
Dinesh Ramanathan, Onsemi's SVP of corporate strategy, makes his case at the level of the whole rack. "The space in the rack isn't changing, but the amount of power that rack is going to end up consuming is increasing by a dramatic amount," he said [4]. He also said: "If we can't increase the power density across the entire rack, we'll never be able to reach 1MW" [5]. His own numbers size that job. From 60kW to 1MW is 16.7 times the power in the same space, and the first leg, 60kW to 200kW, is 3.3 times of it inside two years [1][2].
The method is to stop treating the package as a separate thing. Onsemi cuts a cavity into a silicon wafer and places the die inside it, Ramanathan said, so silicon, silicon carbide and vertical gallium nitride can be interconnected in what the company calls a "highly integrated wafer-level architecture" [6]. FETs, drivers and controllers can sit in one package co-optimized for electrical, thermal and mechanical performance [7]. "By reducing packaging overhead and using the full EPP footprint to conduct heat, EPP can support more compact power systems, improve thermal management and enable greater power density in AI infrastructure," the company said in a statement [10]. It runs on Onsemi's standard 12-inch silicon wafer lines [9].
The customer a buyer can check is an automaker. Subaru has a deployment partnership giving it early access to engineering samples, simulation models and technical expertise [11]. Ramanathan said Onsemi is also working with an industrial company and an AI data center company, and that he was not in a position to name them [12]. He said Onsemi has shown those customers what the technology can do and is working towards getting the product into revenue in the 2027 time frame, so it is sampling customers this year [13]. Qualification is expected in Q1, followed by general availability [14].
Two questions sort any power-density pitch into something you can act on. Where in the power path does the quantified saving land: at protection, at conversion, or at the point of load. And who measured it: the vendor in simulation, or a customer in a deployed rack. EPP's published number sits at protection, measured by Onsemi, in a design the company itself calls early [8]. That is where a new architecture normally starts. For anyone taking samples before the Q1 qualification, the useful ask is how many of a rack's watts pass through EPP silicon and how many millimetres that frees up, because the 60kW-to-200kW step gets engineered before EPP reaches revenue [3][13].
What to watch
- Whether Onsemi names its AI data center partner when EPP is qualified in Q1.
- Any rack-level power density number for EPP published by Onsemi or by a deployed customer.
- Whether the 200kW rack generation ships with EPP silicon inside it or with existing packaging, given the 2027 revenue target.