The new GlobalFoundries process combines photonic and digital components on a single 45nm chip. The primary application is data-center communications, but lidar and computing will benefit as well.
TSMC and Samsung have suffered lengthy delays in their 3nm processes, and the gain in density and other characteristics is smaller than in previous nodes. Intel says its future nodes are on schedule.
The Silicon Valley startup uses a new type of memristor to perform analog math for deep-learning acceleration. It provides nonvolatile storage and performs low-power AI computations.
EdgeCortix launched its first edge-AI inference chip by hardening its DNA IP, delivering low latency and high power efficiency for applications that fit into 5W to 20W.
The new deep-learning accelerator (DLA) can scale to more than 2,000 TOPS, providing a licensable core for applications such as autonomous driving and natural-language processing.
The newest set of MLPerf Inference results showcase the same old vendors; almost all the data-center and edge accelerators came from Nvidia and Qualcomm. Orin was the notable newcomer.
Intel’s new Arc products are its first significant discrete GPUs, using competitive performance and hardware ray tracing to bring new competition to laptop-PC graphics cards.
The dominant AI-chip vendor, Nvidia has raised the bar with its new Hopper and Orin accelerators. Startups, hyperscalers, and large chip vendors try to compete with the company but keep falling short.
Due to sample late this year, Spectrum-4 will quadruple the bandwidth of Spectrum-3 by both doubling per-lane speed to 100Gbps and doubling the number of serdes lanes to 512, yielding 51.2Tbps.
Near-memory and in-memory compute are techniques for reducing computing power—especially for AI. But they mean different things to different companies. Understanding the differences is important for understanding how some AI chips work.
The next-generation AI architecture powers the H100 card and DGX-H100 system. The 700W flagship card triples peak performance over Ampere while adding FP8 support for more-efficient training.
Intel announced its first Agilex to copackage DRAM, sporting a pair of HBM2e stacks that deliver up to 32GB. It’s also using the Intel 7 process to further cut power dissipation.
The R-Car V4H is a new automotive system-on-chip for vehicles with extended ADAS capabilities. It addresses functional safety and the greater AI performance for Level 2+ driving.
In its sixth generation, PCI Express underwent the biggest changes since its debut in 2003, adopting PAM4 signaling. The new modulation enables data rates to reach 64Gbps per lane.
The ADAS-processor leader is developing a single-chip solution for self-driving cars, along with its own lidar and radars. By 2025, Mobileye hopes its system will reduce the car buyer’s add-on cost to just $10,000.
