Senior DFT Engineer - Architecture

Vanja Kovinić joins INSAIT as a Ph.D. Student. September 4, 2026 INSAIT is excited to welcome Vanja Kovinić to INSAIT as a new Ph.D. student! Vanja will pursue his Ph.D. in computer vision under the supervision of Dr. Danda Pani Paudel and Prof. Luc Van Gool. His research interests include 3D Computer Vision, Self-Supervised Learning, and World Models. Before joining INSAIT, Vanja worked as an AI Compiler Engineer at Tenstorrent, where he contributed to an open-source MLIR-based compiler and helped expand model support to 500+ AI models across vision, language, and other domains. He also gained research experience at the University of Illinois Urbana-Champaign, working on mixed-precision quantization. Vanja holds a B.Sc. and M.Sc. in Computer Science from RAF, Union University in Belgrade. Welcome to INSAIT, Vanja! INSAIT is looking forward to having you with INSAIT and following your research journey.

ai& and Tenstorrent launch JapanFold, bringing open-source, sovereign drug discovery to Japan. Japan should never have to choose between world-class AI and data sovereignty. JapanFold removes that trade-off entirely - frontier drug discovery, open-source, sovereign, and fully in-country." - David Bennett, CEO and co-founder, ai& YOKOHAMA, JAPAN, September 2, 2026 /EINPresswire.com/ - ai& and Tenstorrent launched JapanFold, a drug discovery platform running on ai&'s sovereign infrastructure, powered by Tenstorrent Galaxy(TM) superclusters. Japanese researchers, pharmaceutical companies, and biotech firms can now access frontier drug discovery models at a fraction of the cost of conventional cloud infrastructure, with every computation running entirely within Japan. JapanFold serves the leading open-source structural biology models: - Boltz-2, OpenFold3, ESMFold-2, OpenDDE, Protenix-v2, RoseTTAFold3, OpenBind-0, and Nesso-1 for structure and binding affinity prediction, - BoltzGen, RFdiffusion 3, and PXDesign for de novo drug design, covering the full discovery workflow from protein folding through drug candidate generation, and - ESMC and SaProt for protein embeddings to support downstream tasks ai& optimizes and serves every JapanFold model on Tenstorrent processors, delivering 1:1 accuracy without the cost premium - because the models scale nearly linearly on Tenstorrent hardware, doubling the number of processors doubles output. "Japan's research community should never have to choose between world-class AI and keeping their data sovereign. JapanFold removes that trade-off entirely - frontier drug discovery, open-source, sovereign, and fully in-country," said David Bennett, CEO and co-founder, ai&. JapanFold is accessible in three ways: - Workbench: Fold and design proteins directly in the browser. No install, no sign-up. - API: One HTTP endpoint for every model at api.japanfold.aiand.com/. - Agent Skill: Plain-language access from Claude Code, Cursor, Codex, and more. "Scalable hardware and open-source models are changing what's possible in biology and drug discovery," said Jim Keller, CEO of Tenstorrent. "Tenstorrent delivers fast, scalable compute, and ai& owns the full stack in Japan. Together, we've turned it into a platform researchers can actually use with JapanFold." The launch of JapanFold is the first step in ai& and Tenstorrent's broader effort to make sovereign, open-source AI infrastructure available to scientific research communities across the region. About ai& ai& is a global AI technology company founded on the simple conviction that whoever owns and optimizes the full stack wins. By integrating next-generation data center infrastructure, heterogeneous compute, and frontier model services into a single optimized platform, ai& gives enterprises and developers the performance, economics, and data sovereignty that no single-layer provider can match. Founded in Japan and expanding globally, ai& is building the foundation for the AI-native future. About Tenstorrent Tenstorrent is an AI compute company led by CEO Jim Keller - architect of Apple A4/A5, AMD Zen, and Tesla's Full Self-Driving chip. The company builds RISC-V-based AI processors and systems for developers, enterprises, and sovereign infrastructure worldwide. In addition to servers and workstations, Tenstorrent licenses its TT-Ascalon RISC-V CPU and Tensix AI cores to chip designers including Samsung and LG. Backed by Bezos Expeditions, Samsung, LG Electronics, Hyundai Motor Group, Fidelity, and others, Tenstorrent has raised over $1B+ and operates from Santa Clara, Austin, Toronto, Belgrade, Tokyo, and Bangalore. Legal Disclaimer: EIN Presswire provides this news content "as is" without warranty of any kind. Healthcare Industry Today do not accept any responsibility or liability for the accuracy, content, images, videos, licenses, completeness, legality, or reliability of the information contained in this article. If you have any complaints or copyright issues related to this article, kindly contact the author above.

SanDisk and SK hynix release first OCP specification for High Bandwidth Flash for AI Memory wall. August 5, 2026 As the global artificial intelligence economy shifts rapidly from model training to large-scale inference deployment, the semiconductor industry faces an acute physical challenge known as the memory wall. Modern large language models (LLMs) demand immense bandwidth positioned close to compute cores to process tokens efficiently. However, while High Bandwidth Memory (HBM) offers extraordinary speed, its capacity limitations and high costs create severe economic barriers for hyperscale AI deployment. Conversely, conventional NVMe Solid State Drives (SSDs) deliver massive storage density but suffer from throughput bottlenecks that stall high-speed inference processing. To eliminate this memory tier gap, flash memory pioneer SanDisk Corporation and memory leader SK hynix Inc. announced the release of the first official High Bandwidth Flash (HBF(TM) technical specification through the Open Compute Project (OCP). Published six months after the OCP HBF workstream was established in February, the open specification establishes a technical framework for integrating high-density 3D NAND flash directly into near-compute memory hierarchies. Hyperscale AI giant Google and AI processor innovator Tenstorrent joined the alliance as core consortium members, validating the specification against real-world data center workloads. Discover more Social Sciences Engineering & Technology Agriculture & Forestry A Universal blueprint for High Bandwidth Flash. The OCP HBF specification provides chip designers and data center operators with an open framework to integrate HBF alongside HBM and conventional DRAM. By establishing electrical, physical, and protocol guidelines, the alliance enables AI accelerator designers to build flexible systems that increase near-compute memory capacity while lowering total cost of ownership (TCO). Machine Learning & Artificial Intelligence Key technical highlights of the open OCP HBF specification include: UCIe Interconnect Standardization: Utilizes the Universal Chiplet Interconnect Express (UCIe) interface, allowing HBF modules to connect seamlessly with GPUs, CPUs, and custom ASICs. Massive Near-Compute Density: Defines 8-layer and 16-layer 3D NAND stacking configurations delivering up to 512GB of capacity per stack - providing up to 16x the capacity of HBM stacks at lower costs. Multi-Grade Bandwidth Performance: Establishes three performance tiers delivering data throughput from 0.4 TB/s up to 3.0 TB/s, bringing flash storage into near-memory performance thresholds. Open Ecosystem Alignment: Published under the Open Compute Project framework to position HBF as the de facto standard across data centers, preventing vendor lock-in. "AI inference is creating a new set of memory requirements, and HBF technology is designed to meet that moment," stated Alper Ilkbahar, Chief Technology Officer at SanDisk. Engineering & Technology Impact on the semiconductor industry. The release of the HBF specification by SanDisk, SK hynix, Google, and Tenstorrent marks a major structural evolution across the Semiconductors sector: 1. Redefining the Global AI Memory Hierarchy Historically, the memory hierarchy was divided between volatile DRAM/HBM and non-volatile NAND flash drives. The standardization of HBF establishes Tiered Near-Compute Storage. Positioning 3D NAND within a high-speed UCIe interconnect envelope creates an intermediate layer that absorbs massive LLM parameter weights, reducing expensive HBM capacity demands while eliminating PCIe storage bus bottlenecks. 2. Accelerating UCIe Chiplet Ecosystem Adoption Integrating HBF technology via UCIe host interfaces provides a commercial catalyst for advanced packaging and chiplet integration. As memory makers stack 375-layer 4D NAND dies alongside logic dies, the supply chain will see increased capital expenditure toward advanced silicon interposers and 3D chiplet packaging. Overall effects on businesses operating in the sector. For hyperscale cloud providers, AI silicon startups, and enterprise OEMs, the standardized HBF platform delivers direct strategic benefits: Lowering Capital Expenditure for AI Deployments: Integrating lower-cost 3D NAND into near-compute tiers allows cloud providers like Google to serve multi-billion parameter LLMs at lower capital costs per query. Leveling the Playing Field for Startups: Open UCIe-compliant HBF specifications enable non-traditional chipmakers (such as Tenstorrent) to design high-performance AI inference accelerators without building custom memory controllers. Creating New Revenue Streams for Foundries: The emergence of HBF provides NAND flash manufacturers with a high-margin product category, reducing exposure to consumer SSD price cycles. Conclusion. The release of the first OCP High Bandwidth Flash specification by SanDisk and SK hynix represents a defining moment for semiconductor architecture. By combining high-density 3D NAND flash with near-memory bandwidth and open UCIe chiplet connectivity, these industry leaders are dismantling the AI memory wall. For the semiconductor industry, this milestone confirms that the future of scalable AI computing belongs to open, tiered memory architectures capable of delivering massive intelligence at scale.