Intel’s new APX extension adds registers and other features to the instruction set, and AVX10 will enable the company’s little CPUs to support the same vector-processing operations as its performance cores

Yet another South Korean AI chip startup, Deepx, has emerged from stealth mode. The company’s products target computer-vision applications and provide performance from 2 to 184 TOPS.

Neo’s 3D X DRAM technology promises an 8× increase in DRAM capacity with a per-bit cost one-sixth that of conventional DRAM. The company seeks a partner for silicon implementation.

TechInsights’ floorplan analysis reveals MediaTek and Qualcomm have increased GPU size in their flagship smartphone processors while Apple has held its GPU size constant.

At the recent VLSI Symposia, Intel presented three papers discussing PowerVia technology: one on the technology itself, another on the results from a test microprocessor, and a third looking at the possible evolution of PowerVia.

Bluespec’s licensable RISC-V MCUX CPU IP adds custom instructions to its existing ultralow-footprint CPU, branded MCU. Optimized for FPGAs, these CPUs occupy fewer lookup tables than any other commercial RISC-V core.

Tesla has started deploying its Dojo supercomputer that employs unusual technologies to pack AI-training performance in a smaller volume than Nvidia-based systems.

NXP’s i.MX91 addresses simple edge applications with a feature subset of the i.MX93, maintaining pin compatibility and reducing power.

TechInsights’ floorplan analyses of smartphone processors reveal the size of CPUs and caches.

Qualcomm’s Snapdragon 4 Gen 2 offers more-powerful CPUs and faster DRAM than MediaTek’s Dimensity 6100+, but neither new chip delivers much improvement to low-cost smartphones.

Intel’s new accounting approach gives financial responsibility to manufacturing, treating business units as internal customers. This greater accountability should help the overall profitability and competitiveness of its foundry business.

Microcontrollers from STMicroelectronics feature prominently in MLPerf Tiny v1.1 results, and Plumerai’s inference engine demonstrates how it speeds up model execution.