Essential Chips for AI: from the cloud to the edge (2min read)

Image File

Original HTML

Thank you, CEA-LETI, for the opportunity to provide a talk on GF’s portfolio of essential chips that address some of the most pressing semiconductor challenges for AI. Together, we are building a synergistic cloud-to-edge ecosystem to create more powerful AI learning and inferencing and make AI more productive in applications such as autonomous driving, healthcare, robotics, and industrial automation.

In the cloud, connecting more GPUs for more powerful AI models needs higher data rate Fotonix(TM) (silicon photonics) and optical networking; data centers with sustainable power and reduced CO2 emissions requires low transmission loss power delivery and fast transient response voltage regulation from GaN, BCDLite(TM), and FinFET technologies; and more powerful domain specific computing architectures requires chiplet assembly and advanced packaging from GF’s new Advanced Packaging and Photonics Center.

Between the cloud and the edge, higher data speed networks that allow AI to process large data sets quickly and efficiently are made possible with high performance RF such as SiGe, RF-SOI, and FDX(TM) (FD-SOI); massive device connectivity for edge nodes and sensors requires ultra-low-power wireless connectivity from FDX and FinFET; and distributed intelligence, where AI algorithms autonomously process across multiple systems, domains and devices, requires reliable communication and support for new bands and standards such as 5G FR2/FR3, 6G, WiFi8/9, V2X/A2X, SatComm, and THz.

And at the edge, more personalized AI user experiences, where devices tailor their recommendations based on individual preferences and user patterns and provide a higher level of privacy and security by processing data locally requires high-density embedded non-volatile memory (eFlash, eRRAM, eMRAM) for neural net weight storage; battery powered edge computing with real-time inferencing and autonomous decision making at low latency needs ultra-low power FDX (FD-SOI) and FinFET and design IP optimized for high TOPS/Watt; and the fuel for AI data analytics come from Feature-rich sensor solutions including cameras, infrared sensors, microphones, and radars.