Week In Review: Design, Low Power

Tools & IP
MIPS announced its first products based on the RISC-V ISA. The eVocore IP cores are designed to provide a flexible foundation for heterogeneous compute, supporting combinations of eVocore processors as well as other accelerators, with a Coherence Manager that maintains L2 cache and system-level coherency between all cores, main memory, and I/O devices. They target high-performance, real-time compute applications such as networking, data center, and automotive.

Real Intent updated its multimode DFT static sign-off tool, enhancing its fault coverage option that estimates the fault coverage for each test mode, with an overall scan test fault coverage estimation roll-up. It also made additions to the tool’s fine-grained rules, including specialized rules for sequential capture through loops and deep sequences of flip-flops without scan collars, controllability and observability through memories, and advanced connectivity checks.

Faraday Technology made its Gigabit Ethernet PHY available on UMC’s 28HPC+ process. The PHY features both 1000BASE-T copper medium access and 100BASE-FX optical medium access and can implement a 3.3V GbE PHY utilizing 1.8V IO.

Arteris IP reported first quarter 2022 financial results with revenue of $11.8 million, up 77{64d42ef84185fe650eef13e078a399812999bbd8b8ee84343ab535e62a252847} from $6.7 million in the first quarter of 2021. “We’re excited to report a strong start to 2022, with Annual Contract Value plus Trailing Twelve Month Royalties of $52.8 million, up 26{64d42ef84185fe650eef13e078a399812999bbd8b8ee84343ab535e62a252847} year-over-year,” said K. Charles Janac, president and CEO of Arteris IP. “The ongoing democratization of SoC design as well as a disintermediation of the semiconductor supply chain is driving a strong need for automation of System IP solutions in order to compensate for a shortage of SoC architects and skilled interconnect IP engineers.”

Arm announced full year 2021 financial results with revenue of $2.7 billion, up 35{64d42ef84185fe650eef13e078a399812999bbd8b8ee84343ab535e62a252847} compared to the previous year. “Our priority is to continue to deliver on our business strategy, enable partners with the solutions they need through further investment in our roadmaps and engineering talent, and together with our ecosystem redefine the future of computing,” said Rene Haas, Arm CEO.

AMD’s Xilinx Zynq UltraScale+ RFSoC was used in the development of multiple Evenstar radio units (RUs). The Evenstar program led by Meta Connectivity is a collaborative initiative between operators and technology partners to build adaptable, efficient, and metaverse-ready radio access network (RAN) reference designs for 4G and 5G networks in the Open RAN ecosystem.

AI hardware
Tachyum debuted a processor that combines the functionality of a CPU, GPU, and TPU in a homogeneous architecture. The Prodigy processor incorporates 128 high-performance unified 64-bit cores running up to 5.7 GHz, 16 DDR5 memory controllers, and 64 PCIe 5.0 lanes. It handles matrix and vector processing with support for a range of data types, 2×1024-bit vector units per core, AI sparsity and super-sparsity support, and no penalty for misaligned vector loads or stores when crossing cache lines. It targets cloud, AI training and inference, and HPC applications.

Intel’s Habana Labs debuted the second generation of its deep learning processors, both implemented in 7nm. The Gaudi2 Training processor features 24 Tensor Processor Cores and integrates media processing on-chip, 96 GB on-board memory and 48 MB SRAM, providing up to three times the training throughput over first-generation. On the inference side, the Greco Inference processor integrates media encoding and processing on-chip, supporting media formats HEVC, H.264, JPEG and P-JPEG, as well as support for multiple data types.

Data center, HPC
Marvell will acquire Tanzanite Silicon Solutions, a developer of Compute Express Link (CXL) technologies. It offers an SoC that enables independent scaling and sharing of memory and compute in a pool with low latency within and across server racks. “The CXL standard will play a significant role in helping the industry deliver fully composable infrastructure for the cloud,” said Noam Mizrahi, corporate chief technology officer, Marvell. “The integration of CXL across our end-to-end, cloud-optimized silicon portfolio will bring new levels of data center efficiency, scalability and flexibility to power emerging metaverse and next generation AI applications.” Tanzanite Silicon Solutions is based in Milpitas, California, and was founded in 2020. Terms of the all-cash deal were not disclosed.

Samsung Electronics introduced a 512GB Compute Express Link (CXL) DDR5 DRAM module. “CXL DRAM will become a critical turning point for future computing structures by substantially advancing artificial intelligence (AI) and big data services, as we aggressively expand its usage in next-generation memory architectures including software-defined memory (SDM),” said Cheolmin Park, vice president of Memory Global Sales & Marketing at Samsung Electronics, and Director of the CXL Consortium.

Montage Technology uncorked its Compute Express Link (CXL) Memory eXpander Controller. The device is designed to be used in add-in cards (AIC), backplanes, or EDSFF memory modules to provide a high-bandwidth and low-latency interconnect between the CPU and the CXL-based devices for data-intensive applications such as HPC and AI. It uses CXL 2.0 and supports PCIe 5.0, DDR4, and DDR5.

Western Digital announced a slew of new flash and HDD products for cloud, client, and consumer applications. Among the new products are 22TB CMR and 26TB UltraSMR HDDs for data centers. The  energy-assisted PMR (ePMR) drives use the company’s OptiNAND technology that includes an iNAND universal flash storage (UFS) embedded flash drive (EFD). The UltraSMR technology introduces large block encoding along with an advanced error correction algorithm that increases tracks-per-inch (TPI) to enable higher capacity. It also introduced a new line of high-capacity NVMe PCIe 4.0 SSDs for cloud environments.

Arm made its toolchain package for HPC application developer available free of charge. Arm Compiler for Linux includes LLVM-based C/C++/Fortran compilers and optimized libraries with BLAS, FFT, LAPACK, Sparse Linear Algebra functionality.

Power devices
Keysight Technologies announced its next-generation Double-Pulse Tester (DPT) for faster testing of entire power modules. PD1550A Advanced Dynamic Power Device Analyzer can test modules up to 1360 V and 1000 A. “The new PD1550A expands our offering into whole power module testing, which is used in high-power converter designs, while maintaining ease-of-use and compliance to worldwide safety regulations,” said Thomas Goetzl, vice president and general manager for Keysight’s automotive & energy solutions business unit.

Infineon Technologies expanded its CoolSiC portfolio with high-voltage solutions, adding 2 kV silicon carbide (SiC) MOSFETs, along with a 2kV SiC diode for applications up to 1500 V DC. The new SiC MOSFET combines both low-switching losses and high-blocking voltage in one device that can optimally meet the requirements of 1500 V DC systems. The new 2 kV CoolSiC technology offers a low drain-source on resistance (R DS(on)) value. In addition, the rugged body diode is suitable for hard switching. They target next-generation photovoltaic, EV charging, and energy storage systems.

Infineon also launched a new 1700 V IGBT7 chip in the standard industrial package EconoDUAL 3. It provides a leading current of 900 A and 750 A, and enables an improved power range for inverters. The modules target a wide range of applications including wind, drive, and static VAR generators (SVG).

STMicroelectronics uncorked new N-channel super-junction multi-drain silicon power MOSFETs targeting switched-mode power supplies in applications from data-center servers and 5G infrastructure to flat-panel televisions.

Navitas Semiconductor released a new high-power 650/800 V-rated GaN power IC that integrates power, drive, control, with additional autonomous-protection and loss-less current-sensing. It targets higher-power applications such as 400-1000 W 4K/8K TVs and displays, next-generation gaming systems, 500 W solar microinverters, 1.2 kW data center switched-mode power supplies, and up to 4 kW / 5 hp motor drives.

Onsemi debuted a TO-Leadless (TOLL) packaged silicon carbide (SiC) MOSFET. The company says it provides smaller area, better thermal performance, and lower package inductance compared to D2PAK packaging. It targets switched-mode power supplies, server and telecommunication power supplies, solar inverters, uninterruptible power supplies, and energy storage.

Infineon Technologies reported second quarter 2022 financial results with revenue of €3.298 billion (~$3.5 billion), up 22{64d42ef84185fe650eef13e078a399812999bbd8b8ee84343ab535e62a252847} from €2.7 billion (~$2.8 billion) in the same quarter last year. “Infineon continues to perform well within an increasingly challenging environment. Revenue and Segment Result both went up again in the second quarter. Global uncertainties, in particular the war in Ukraine and the further course of the corona­virus pandemic, are placing stress on supply chains. At the same time, demand for our products and solutions continues to exceed supply significantly,” said Jochen Hanebeck, CEO of Infineon.

Quantum computing
IBM debuted its new roadmap for large-scale, practical quantum computing. Key to the roadmap is development of modular architectures and networking that will enable systems with larger qubit counts. The company plans to introduce a universal quantum processor with 1,121 qubits in 2023. By 2025, the company has a goal of a 4,000+ qubit processor built with multiple clusters of modularly scaled processors. Another area of focus will be a software orchestration layer that can efficiently distribute workloads and abstract away infrastructure challenges. “By combining modular quantum processors with classical infrastructure, orchestrated by Qiskit Runtime, we are building a platform that will let users easily build quantum calculations into their workflows and so tackle the essential challenges of our time,” said Jay Gambetta, vice president of Quantum Computing and IBM Fellow. One part of the roadmap is circuit knitting, which gets a more detailed explanation in a separate blog. The technique allows large quantum circuits to be partitioned into subcircuits that fit on smaller devices, incorporating classical simulation to “knit” together the results to achieve the target answer.

ColdQuanta acquired Super.tech, a provider of quantum software application and platform development. “The quantum industry is experiencing tremendous growth as hardware and software solutions begin to take root in the real world,” said ColdQuanta CEO Scott Faris. “Since its launch, Super.tech has been recognized as an innovator in quantum software and benchmarking. The combination of Super.tech’s software applications and algorithms and ColdQuanta’s ground-breaking quantum computer, Hilbert, will enable customers to accelerate time to value.” Super.tech is based in Chicago, Illinois, and was founded in 2020.

ColdQuanta also introduced a gate-based cold atom quantum computer. It supports IBM’s Qiskit API and will initially be available in beta through a multi-tenant cloud platform, with public cloud services introduced later this year.

A new quantum computing startup launched from stealth. Diraq is working to build quantum processors using electron spins in CMOS quantum dots, a technology it claims can be scaled to billions of qubits.

Read the latest Low Power-High Performance and Systems & Design newsletters to find out about changes in embedded software, design considerations for DRAM, optimizing architectures, and more.