Fab investments; silicon vs WBG materials; compressing simulation code; Ansys acquires Diakopto

Competition heats up for processors aimed at cloud computing. Ampere (Santa Clara, CA) has just introduced a new family of what it calls “cloud native processors”, with 192 custom designed Ampere cores and a number of features aimed at cloud usages like AI. More news this week include innovations from both industry and academia.

New fab investments: Analog Devices in Europe, Micron in Japan

Analog Devices will invest €630 million at its European regional headquarters in Limerick, Ireland, to build a new 45,000 sq-ft R&D and manufacturing facility which will focus on signal processing innovations. The new facility is expected to triple ADI’s European wafer production capacity. The investment is planned as part of a collaboration within the European Union’s Important Projects of Common European Interest on Microelectronics and Communication Technologies (IPCEI ME/CT) initiative, and is supported by the Irish Government. One year ago, Analog Devices announced a separate investment of €100 million in “ADI Catalyst” at its Limerick campus. Ireland is also home to ADI’s main European Research and Development Center.

Micron Technology will be introducing EUV lithography to its Hiroshima (Japan) fab, to manufacture its next generation of DRAM, the 1-gamma (1γ) node. The company expects to ramp EUV into production on the 1-gamma node in Japan (as well as in Taiwan) from 2025 onwards. Micron will be the first semiconductor company to bring EUV technology to Japan for production, and expects to invest up to 500 billion yen in 1-gamma process technology over the next few years, with close support from the Japanese government.

iDEAL silicon power devices challenge wide bandgap materials

Fabless startup iDEAL Semiconductor has launched its SuperQ technology for silicon power devices. Leveraging both process and architecture innovations, the company claims that SuperQ achieves improvements in silicon that rival the improvements offered by other materials, but with silicon’s manufacturability, availability, and reliability. For example, SuperQ-based 200V MOSFETs deliver 6x lower resistance than existing silicon and 1.6x lower resistance than gallium nitride. The technology, which uses state-of-the-art CMOS equipment, has been developed through collaborations in the U.S. with Applied Materials and Polar Semiconductor. iDEAL has recently received a Series C funding round bringing total investment to over $75 million. As explained on the company website, SuperQ is based on an “asymmetrical charge-balanced trench”: legacy power device architectures have practical limits that bound the N-Conduction region to 50% of the overall structure. The remaining 50% is used for voltage blocking and does not support conduction. iDEAL has invented a charge-balancing method that can be as thin as 5% of the total structure. This opens more room for conduction and drives efficiency higher. This trench technology also allows higher doping concentration in the conduction region due to its blocking efficiency, thus reducing the resistance of the channel and lowering power loss. “We’ve had this drum beat for the past ten years that silicon is dead,” said Mark Granahan, iDEAL’s chief executive officer, interviewed by Reuters. “It’s quite frankly the lack of innovation in silicon that has enabled that drum beat.” Granahan reportedly said silicon carbide can still beat iDEAL’s SuperQ for certain applications including the extremely high voltage devices, but SuperQ could be competitive for about 90% of the overall market for power devices.

Matricization of HPC algorithms achieves 150x simulation speed on Graphcore processors

A research team from King Abdullah University of Science and Technology (Saudi Arabia) has demonstrated execution rates with speedup factors up to 150X/14X/25X/40X, respectively, on applications in computational astronomy, seismic imaging, wireless communications, and climate/weather predictions, running re-structured HPC algorithms on Graphcore Wafer-on-Wafer Intelligence Processing Units. The main concept of the work is that legacy High Performance Computing applications can leverage the new AI acceleration processors after an algorithmic redesign based on the “matricization” technique: exploiting data sparsity via algebraic compression; and expressing the critical computational phases in terms of tile low-rank matrix-vector multiplications (TLR-MVM) and batch matrix-matrix multiplications (batch GEMM). Algebraic compression enables to reduce memory footprint and to fit into small local cache/memory, while batch execution ensures high occupancy. “We can perform compression on the matrix operator that describes the physics of the problem, while maintaining a satisfactory accuracy level as if no compression was done,” said the team leader Hatem Ltaief in an interview.

Retina-inspired image sensor based on RGB perovskite photodetectors

A research team from Pennsylvania State University has developed an image sensor inspired by the retina, the light-sensitive part of human eye. The sensor consists of an array of three different types of narrowband perovskite photodetectors, each type only sensitive to red, green or blue light. In contrast, conventional charge-coupled device and CMOS cameras use broadband sensor arrays that cannot distinguish the color and usually need an external color filter array (CFA) to achieve monochromatic sensing. The insertion of CFA not only increases the cost and manufacturing complexity but also causes spatial information loss and other problems. The retina-inspired sensor also includes a neuromorphic algorithm, mimicking the preprocessing performed by human eye. Additionally, since perovskite generate power as it absorbs light, the sensor enables a power-free photodetection feature.

Acquisitions

Ansys has entered into a definitive agreement to acquire Diakopto, an EDA startup focusing on critical issues caused by interconnect parasitics. According to Ansys, Diakopto’s products provide actionable analytics to guide designers to fix these problems – a capability that has not existed before. Through early identification and what-if analysis of parasitic problems, engineers can minimize costly iterations late in the design cycle.

Infineon has acquired Sweden-based startup Imagimob, a platform provider for machine learning solutions for edge devices. Imagimob specializes in Tiny Machine Learning and Automated Machine Learning (AutoML) solutions, providing an end-to-end development platform for machine learning on edge devices. Imagimob platform’s use cases include audio event detection, voice control, predictive maintenance, gesture recognition, signal classification as well as material detection.

This entry was posted on Friday, May 19th, 2023 at 2:08 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.