AI inference accelerator bolsters efficiency in power modules
Power modules for data centers are incorporating AI inference for applications such as agentic AI, response generation with large language models (LLMs), and predictive analytics in finance and healthcare. The use of AI accelerators is mainly aimed at boosting energy efficiency in high-density boards.
Take the case of Infineon, which is incorporating d-Matrix’s Corsair inference accelerator in its OptiMOS TDM2254xx dual-phase power modules. According to Sid Sheth, founder and CEO of d-Matrix, Corsair was purpose-built for delivering the sub-2 ms token latency that interactive applications require.
The OptiMOS TDM2254xx dual-phase power module enables vertical power delivery while offering a density of 1.0 A/mm2. Source: Infineon
Infineon has been working closely with d-Matrix to optimize the Corsair inference accelerator for its power semiconductors. “Infineon has been collaborating with customers specializing in inference processors, such as d-Matrix, from the early days when the industry was mostly focused on training hardware,” said Raj Khattoi, VP and GM of consumer, computing and communication at Infineon.
Infineon, which offers a broad portfolio of power semiconductors, based on silicon (Si), silicon carbide (SiC), and gallium nitride (GaN), has also been working closely with AI companies in both the training and inference markets. And these liaisons have aimed to improve energy efficiency at higher power density in hardware at data centers and other AI installations.
Related Content
- Solving AI’s Power Struggle
- TI launches power management devices for AI computing
- Taiwan’s Emerging Power Electronics Strategy in the AI Era
- Why AI Is Redefining the Future of Commercial Power Infrastructure
- Power Module Packaging Evolves as Materials and Supply Chains Redefine Power Electronics
The post AI inference accelerator bolsters efficiency in power modules appeared first on EDN.
