Week In Review: Design, Low Power

Tools & IP
Synopsys unveiled a new neural processing unit (NPU) IP and toolchain. DesignWare ARC NPX6 NPU IP scales from 4K to 96K MACs with power efficiency of 30 TOPS/Watt. A single instance offers 250 TOPS at 1.3 GHz on 5nm processes in worst-case conditions, or up to 440 TOPS by using new sparsity features, which can increase the performance and decrease energy demands of executing a neural network. It offers optional 16-bit floating point support inside the neural processing hardware. The accompanying development toolkit automatically compiles and optimizes models for neural networks such as CNNs and RNNs/LSTMs, as well as emerging deep learning networks such as transformers and recommenders. A version of the IP is also available for safety-critical automotive SoCs.

Cadence launched a suite of computational fluid dynamics (CFD) solutions for multiple vertical markets, including automotive, turbomachinery, marine, and aerospace. The company says the next-generation high-order flow solver in Fidelity CFD delivers up to 10X the accuracy of standard flow solvers. It offers several specialized flow solvers for marine and turbomachinery applications, as well as general-purpose flow solvers for a broad range of flow types.

BAE Systems’ FAST Labs research and development organization will use Movellus’ Intelligent Clock Network IP with its rad-hard ASIC technologies and libraries to deliver a complete, advanced development solution for creating SoC designs for its portfolio of rad-hard electronics for civil, commercial, and national security missions.

CoreHW debuted new high-current 5A DCDC IP. The DCDC IP block contains most buck converter parts, including power stage with drivers, bootstrap voltage generator, control logic, reference and error amplifiers, fast comparator, under- and overvoltage detection circuitry, over-current protection, current monitoring, 10-bit DAC for output voltage setting and zero-cross detector.

MosChip Technologies uncorked enhanced simplex High Speed Serial Trace Probe (HSSTP) PHY macro with link layer supporting data transfer capabilities of up-to 12.5Gbps per lane in 6nm FinFET technology.

AI hardware
EdgeCortix debuted its Sakura AI co-processor. It uses the company’s 40 TOPS single core Dynamic Neural Accelerator (DNA) IP and can run multiple deep neural network models together with ultra-low latency. The DNA IP is specifically optimized for inference with streaming and high-resolution data. Built on TSMC’s 12nm finFET process, it will initially be available as low-power PCIe based development boards.

Memory
OpenEdges is preparing to go public on the Korean KOSDAQ exchange. The IPO is expected to take place in the third quarter of 2022. OpenEdges is a supplier of IP with a portfolio including DDR memory controller, DDR PHY, NPU, and on-chip interconnect solutions.

Winbond Electronics and Infineon Technologies introduced new higher bandwidth HYPERRAM 3.0. Targeting low power, space-constrained IoT applications that require an off-chip external RAM, HYPERRAM 3.0 operates at a maximum frequency of 200MHz with a 1.8V operation voltage, with a data-transfer rate of 800MBps, doubling the previous generation. It operates via an expanded IO HyperBus interface with 22 pins. 256Mb devices are now sampling.

Wireless, networking, cloud
Mobix Labs will acquire Advanced Interconnect Technologies, a manufacturer and product development firm serving the military, defense, and commercial air and space industries. Mobix Labs develops CMOS-based mmWave beamformers, antenna solutions and RF/mixed signal semiconductors as well as hybrid active optical cables, transceivers, and optical engines. “We are very excited about adding Advanced Interconnect Technologies to our growing company,” said Fabian Battaglia, CEO of Mobix Labs. “Bringing AIT into the Mobix Labs portfolio as part of our acquisition strategy allows us to offer manufacturing and design services to existing global customers, and also opens new doors to deliver our wireless and connectivity products to even more clients in the defense, air and space industries.”

Xiaomi selected Keysight’s automated field-to-lab device test platform solutions to validate the performance of 5G devices, including smartphones and IoT devices, under a wide range of network signaling and radio channel conditions.

Oracle deployed 3rd Gen AMD EPYC processors to power the new Oracle Cloud Infrastructure (OCI) E4 Dense instances, which are part of the Oracle Cloud VMware Solution used to build and run a hybrid-cloud environment for VMware based workloads.

Automotive
Keysight debuted an oscilloscope-based automotive protocol trigger/decode solution covering CAN XL (Controller Area Network eXtra Long) for verification and debugging of low-speed automotive serial bus protocols and simplification of the development and troubleshooting of systems including CAN/CAN FD (Flexible Data Rate) and CAN XL. Keysight’s automotive decoder software triggers and decodes the SIC mode signal in the arbitration phase, as well as the fast mode signal in the data phase, correctly. It also handles the CAN XL protocols implemented with the CAN HS/FD/SIC transceivers.

Find more of this week’s news at Manufacturing, Test and Auto, Security, Pervasive Computing.

Explore the latest Low Power-High Performance and Systems & Design newsletters.