onsemi unveiled its Embedded Power Platform architecture that utilizes the silicon wafer as an embedded package to dramatically increase power density for AI racks, automotive, and industrial deployments.
At its investor day today, onsemi introduced the Embedded Power Platform (EPP), a power-integration approach that embeds heterogeneous power dies directly into silicon and interconnects them through wafer-level redistribution layers rather than traditional wire bonds. Positioned for vehicle electrification and AI power infrastructure, EPP addresses a critical industry challenge: as power density becomes a hard constraint, the conventional practice of optimizing electrical, thermal, and mechanical designs separately has reached its limits. This integrated methodology was a key point emphasized by onsemi during a call with us.
onsemi frames the launch around escalating power demands. Across AI, automotive, and industrial markets, designers must deliver more power in less space while simultaneously improving efficiency and managing heat, size, and cost. When electrical, mechanical, and thermal systems are tuned independently, advancements in one area inevitably create trade-offs elsewhere, forcing costly late-stage design revisions.
EPP is built around a silicon embedded package where heterogeneous power dies sit directly within the substrate and connect via wafer-level metallization. This replaces wire-bonded interconnects with precision redistribution layers, bringing semiconductor-fab precision to the connections and lowering parasitic inductance for superior device control and higher switching frequencies. The platform is technology-agnostic and scales across various semiconductor materials and power levels, supporting silicon, SiC, GaN, and future technologies in configurable combinations. Drivers and controllers are integrated directly with the power devices to reduce system-level complexity. The architecture leverages onsemi’s proven fabrication infrastructure in upstate New York, located not far from where we produced our IBM Z17 video last year, alongside advanced simulation capabilities. The company states this infrastructure shortens time to market while delivering high-voltage isolation and a lower-resistance thermal path.
The platform delivers significant thermal and electrical improvements. Full-footprint thermal conduction enhances heat dissipation and supports higher continuous power operation, while integrated high-voltage isolation reduces reliance on thermally inefficient insulating materials. Shorter, more controlled electrical paths further minimize parasitic inductance. Additionally, digital twin simulation paired with multi-physics co-optimization allows engineers to iterate designs rapidly, reducing physical hardware builds and accelerating development cycles.
onsemi is initially targeting vehicle electrification and AI power infrastructure. In automotive applications, reduced inductance enables faster switching, while full-footprint thermal conduction facilitates smaller, lighter traction inverters that scale across multiple vehicle platforms. For AI data centers, higher power density frees board and rack space, enabling thermally efficient, compact power delivery. The architecture also extends to fast-charging infrastructure, energy storage, grid modernization, and industrial automation. To advance automotive adoption, Subaru has signed on as the first strategic technology engagement partner. onsemi is providing Subaru with early access to engineering samples, simulation models, and technical expertise to evaluate whether integrated power design can improve efficiency and simplify development for next-generation vehicle architectures. Given the longer design cycles typical of the automotive sector, this partnership serves as a foundational customer evaluation phase.
Regarding the AI data center, we asked about primary applications, and the company outlined a particularly compelling use case. Rather than relying on bulky electromechanical breakers for a several-hundred-kW 800V rack, the EPP architecture enables a significantly smaller digital breaker solution. The company advised us to picture replacing the traditional large breaker with this compact alternative, which is easily liquid-cooled or air-cooled, fully controllable, and capable of outputting telemetry data from the assembly. As noted, the first major commercial use case will likely target these 800V DC AI racks.
For readers focused on AI infrastructure, EPP addresses a distinct layer of the power delivery challenge. While recent industry coverage has highlighted solutions like Wiwynn and TE Connectivity’s liquid-cooled 800V DC busbars designed to route power through racks, EPP tackles the problem closer to the source by integrating power devices, interconnects, and control functions within a single package. As AI racks continue to grow denser, optimization for next-generation deployments is expanding beyond GPUs, NICs, CPUs, switches, storage, and liquid cooling to encompass power delivery itself. Notably, the 800V platform standard is being leveraged across both automotive and AI data center ecosystems. On a final note, onsemi’s headquarters is located approximately ten minutes from STH’s studio.