Export of some Nvidia and AMD products to China halted; Intel’s Risc-V IDE; AI chip reaches 30 TFlops/W

More U.S. and western European tech companies have reportedly closed their Russian operations: among them Dell, Logitech, Ericsson and Nokia. Another significant update on geopolitical matters is the export ban on some Nvidia and AMD products (see below). However, the effectiveness of sanctions against China is a debated issue – see, for example, this EETimes article – and reverse engineering on a SMIC chip has provided additional surprises: TechInsights has found many similarities in process technologies, designs and innovations between SMIC’s 7-nanometer and TSMC’s 7-nanometer nodes. According to TechInsights, also, it is a notable achievement for SMIC having moved from 14-nanometer to 7-nanometer in just two years, without access to the most advanced western equipment and technologies.

Export restrictions on some advanced Nvidia GPUs and AMD accelerators

Nvidia and AMD have reportedly been told by the US government to halt exports of certain high-performance chips and systems to China. As for Nvidia, the restrictions cover A100 and forthcoming H100 GPUs, and any systems that include them, effective immediately. AMD has reportedly been given new requirements by the US Department of Commerce that will hit shipments of its MI250 accelerator to China. In a regulatory filing, Nvidia said that the export restrictions are due to a potential risk of the products being used by, or diverted to, a “military end user.” Both companies said the new mandate also covers a ban in export to Russia.

Intel’s Risc-V integrated development environment

Intel has launched its own integrated development environment for Risc-V, targeted at pre-silicon development, allowing for a variety of Risc-V cores and other IP to be instantiated on Intel FPGA and simulator platforms. Called Pathfinder, the solution offers the ability to run industry leading operating systems and tool chains within a unified IDE. Intel Pathfinder is initially available in two versions: the Starter Edition, for the hobbyist, academia and research community, available as a free download; and the Professional Edition, targeting organizations involved in commercial Risc-V based silicon and software. Intel Pathfinder leverages contributions from several ecosystem partners: Risc-V cores from Andes, Chips Alliance, Codasip, Fraunhofer IMS, MIPS, OpenHW Group, and SiFive; other IPs from Cadence (Tensilica DSPs) and STMicroelectronics (LSM6DSOX inertial module); cybersecurity solutions from Check Point Software Technologies (Quantum IoT Protect Nano Agent) and Crypto Quantique; various tools from Codeplay, Imperas, IOTech and Siemens EDA. More Pathfinder ecosystem members include SoC.one (cloud platform) and Terasic (FPGA boards).

Pathfinder is just the latest in a series of Risc-V focused initiatives recently taken by Intel under the lead of CEO Pat Gelsinger. Risc-V is at the heart of the new Nios V soft processor for Intel FPGAs, and is also used in the new Mobileye EyeQ Ultra automotive platform; Intel Foundry Services, too, has announced several partnerships and investments supporting the Risc-V architecture.

Alibaba’s Risc-V development platform

More news concerning the development of Risc-V-based SoCs are coming from China. Alibaba Cloud has unveiled a development platform named Wujian 600, aimed at edge-AI computing. With the new platform, developers can run desktop-level applications, such as FireFox browser and LibreOffice built on OpenAnolis, an open-source Linux-based operating system launched by Alibaba in 2020.

Keysight’s device modeling 2023 software suite

According to Keysight, the 2023 edition of the company’s device modeling software suite offers workflow and speed improvements to the engineers who generate high-quality SPICE models and process design kits for baseband and RF IC designs that leverage both silicon and compound III-V technologies. Improvements include a new modeling flow manager, a new link to Synopsys’ PrimeSim HSPICE, an advanced Low-frequency Noise Analyzer supporting M9601A PXIe Precision Source/Measure Unit, and more.

New model generator in PathWave Device Modeling (IC-CAP) 2023. Credit: Keysight

EPFL benchmark results for Rapid Silicon’s Raptor software

FPGA vendor Rapid Silicon has announced that its commercial open-source FPGA EDA suite, Raptor, was awarded 24 verified unique wins and two ties – 2x more wins than the leading competitor- in the latest Combinatorial Benchmark Suite from École Polytechnique Fédérale de Lausanne (EPFL, a Swiss academic institution). The company attributes these results in large part to its “ABC-DE” algorithm. Competitors were commercial synthesis tools, as Rapid Silicon was the only FPGA vendor participating in the EPFL benchmark.

Deep learning acceleration updates: MemryX, GrAI Matter, Biren, Untether AI, SiMa.ai

Catching up on some recent announcements in the deep learning acceleration space, let’s begin with startup MemryX that is now sampling its MX3 accelerator for edge AI, powered by a proprietary compute-at-memory technology and data-flow architecture. The company claims this is “the easiest-to-use and most efficient edge AI accelerator”, allowing customers to compile any trained AI model with just a single click. Neuromorphic computing company GrAI Matter has reportedly received $1 million in pre–orders for its GrAI VIP chip. On occasion of Hot Chips 2022, Chinese startup Biren Technology emerged from stealth and unveiled its new general-purpose GPU chip BR100, aimed at datacenter applications. The device consists of two 7-nanometer chiplets manufactured by TSMC and packaged using CoWoS technology. Biren has also invented its own number format, which it calls TF32+. Also at Hot Chips, Untether AI announced its new architecture based on ‘at-memory computation’ for accelerating AI inference workloads called speedAI, with the internal codename “Boqueria”, claiming 30 TeraFlops per watt (TFlops/W) and 2 PetaFlops of performance. Lastly, SiMa.ai has started shipping its purpose-built machine learning SoC platform for computer vision applications.

Acquisitions

Renesas has entered into a definitive agreement to acquire Steradian, a fabless semiconductor company based in Bengaluru, India, that provides 4D imaging radar solutions, in an all-cash transaction. Renesas aims to develop complete automotive radar solutions that combine ADAS SoCs for processing radar signals, power management ICs, and timing products together with software for object recognition. Steradian’s radar technology is expected to be adopted also in home security systems, traffic monitoring, gesture recognition etc.

Upcoming events

ARC Processor Summit, September 8, Santa Clara, CA.

4th Panel Level Packaging Consortium Symposium, September 8, Berlin, Germany.

International Conference on Silicon Carbide and Related Materials (ICSCRM), September 11-16, Davos, Switzerland.

Women in Semiconductor Hardware (WISH) Conference, September 13, Santa Clara, CA.

This entry was posted on Friday, September 2nd, 2022 at 11:13 am. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.