SK Hynix

Empowering Tomorrow's Innovations with Generative AI Infrastructure

Generative AI Infrastructure

d-Matrix is a chip design company specializing in the creation of chips optimized for GenAI applications. The company's core innovation lies in its ability to develop chips that significantly reduce energy consumption while managing the substantial computational demands of AI processing.&nbsp;d-Matrix’s first generation of AI hardware products includes the Nighthawk, Jayhawk-I, and Jayhawk II chiplets. These were succeeded by the Corsair C8 compute card, introduced in September 2023. d-Matrix claims the Corsair C8 compute card outperforms NVIDIA’s H100 GPU by a factor of nine in GenAI tasks. This AI accelerator utilizes d-Matrix's proprietary digital-in-memory computer (DIMC) technology, providing ample memory capacity for AI models. d-Matrix claims that a cluster of Corsair C8 cards can deliver 20x better power efficiency, 20x lower latency, and up to 30x lower costs compared to similar NVIDIA solutions.In&nbsp;<a href="https://sp-edge.com/updates/36641">November 2024</a>, announced Corsair, an AI inference computing platform for data centers. The platform comes in a PCIe Gen5 full-height, full-length card form factor and will be broadly available in Q2 2025.Key customers and partnershipsIn November 2022, d-Matrix partnered with Microsoft to utilize Microsoft's Project Bonsai, a low-code reinforcement learning (RL) platform, to develop an AI-trained compiler for d-Matrix's DIMC products.

Founded in Israel, NeuReality specializes in providing hardware infrastructure to streamline computation processes to accelerate the development of AI applications. Its technology assists AI inferencing deployment by offering a comprehensive system solution that seamlessly supports a vast array of deep learning models.NeuRealitys’ infrastructure includes an architecture called the Network Addressable Processing Unit (NAPU) to maximize the capabilities of Deep Learning Accelerators (DLAs). This design facilitates inference through purpose-built NAPUs that are specifically crafted for AI inference tasks, enhancing performance and efficiency.The NR1 AI Inference Solution by NeuReality is a comprehensive system comprising architecture, hardware, and software platforms. Launched in November 2023, it intends to simplify the installation, utilization, and management of AI, making it accessible and manageable for users.The hardware offerings of the solution include,1. NR1 Network Addressable Processing Unit : A network addressable inference Server-on-a-Chip with an embedded Neural Network Engine, which is a workflow-optimized hardware device with specialized processing units, native network capabilities, and virtualization capabilities2. NR1-M™ AI Inference Module : A full-height, double-wide PCIe card containing one NR1 Network Addressable Processing Unit (NAPU) system-on-chip and a network addressable Inference Server that can connect to an external Deep Learning Accelerator (DLA)3. NR1-S AI Inference Appliance : An optimized design for an inference server that contains NR1-M modules with the NR1 NAPU, which enables disaggregated AI service in a scalable and efficient architectureThe software component of the solution is designed to be compatible with any trained model from any development environment. It includes tools that handle the entire AI pipeline and enable seamless connection of AI workflows to any environment.NeuReality differs from traditional AI chip connectivity in replacing CPU-centric support with an AI-centric platform. This allows components to communicate directly with AI accelerators, eliminating the need for commands to pass through a central processing unit. Additionally, the technology enables AI accelerators to connect with each other, enhancing efficiency and performance in AI applications. It also claims its AI-centric approach reduces complexity, cost, and power usage. The firm claims it can reduce inference costs by 90%.Key customers and partnershipsIn November 2021, NeuReality teamed up with IBM to develop advanced AI inference technology by enhancing its NR1 chip and system for use in IBM's Hybrid Cloud.In September 2021, the company partnered with Xilinx to develop AI-centric platforms that improve real-world AI applications. This collaboration was intended to focus on NeuReality's AI-centric inference platform NR1-P and address existing issues with AI deployment including cost and complexity.

Generating, collecting, and analyzing enterprise data will never be the same.

Edge Computing

<div>
 
 Headquartered in Germany, Ferroelectric Memory GmbH (FMC) provides ferroelectric hafnium oxide non-volatile memory chips for semiconductor and fabless manufacturers. This benefits Internet of Things (IoT) devices, including those at the edge, as the chips enable products like microcontrollers (small computers on integrated chips) to improve their speed and capability.
 
 
 The company claims its chips offer fast switching, ultra-low power and scalability enhancements over competitors. The chips are suitable for a range of applications including IoT, consumer electronics, automotive, medical, and aerospace. FMC also works with semiconductor companies and foundries in the US, Europe, and Asia.&nbsp;
 
 
 In November 2020, FMC raised USD 20 million in Series B funding to accelerate the commercialization of its artificial intelligence (AI), IoT, and embedded memory applications, among others. The company also intends to expand the team at its Dresden headquarters and begin expanding internationally to the US and Asian markets.
 
</div>

The toolkit to build machine learning applications at scale

Machine Learning Infrastructure

d-Matrix specializes in the development of computing platforms tailored for managing GenAI inferencing workloads within data centers. It aims to overcome memory limitations prevalent in AI computing by advancing high-efficiency digital-in memory compute (DIMC) chips.&nbsp;Its flagship product Corsair, scheduled for commercial release in&nbsp;<a href="https://sp-edge.com/updates/36641">Q2 2025</a>, serves as an in-memory compute engine engineered to accommodate the entire AI model within the memory, giving developers the flexibility to deploy AI models. Further, the platform offers 2400 TFLOPs of eight-bit peak compute, 2 GB of integrated performance memory, up to 256 GB of off-chip capacity memory, and delivers ultra-high memory bandwidth of 150 Tbps.

A new generation of building blocks for emerging tech

Next-gen Semiconductors

SiFive is a fabless semiconductor company founded in 2015 that specializes in designing processor cores and chips based on the open-source RISC-V instruction set architecture. The company provides commercial RISC-V-based platforms and custom silicon solutions, offering a range of processor core IP, development boards, and SoCs. SiFive's technology enables system designers to quickly and easily build custom chips for various applications including data centers, networking, storage, and edge computing. The company's cores offer 30-40% better power efficiency and a smaller size compared to alternatives, allowing for more cores on a chip. SiFive's online platform simplifies the process of designing and testing chips and core IP, even for those new to silicon development. In September 2020, SiFive appointed Patrick Little as President and CEO to accelerate adoption of its technology for next-generation computing solutions. As of March 2022, SiFive had a post-money valuation between USD 1-10 billion.
Key customers and partnerships
SiFive has collaborated with several high-profile organizations. In 2022, NASA announced that SiFive's RISC-V-compatible CPU cores would power its new High-Performance Spaceflight Computer, expected to deliver a 100-fold improvement in performance over its predecessor. SiFive has also worked with the Barcelona Supercomputing Center to create an open-source API for popular compilers, enabling applications to use the RISC-V Vector Extension for high-performance computing, artificial intelligence, and computer vision. The company has partnerships with major technology firms, as evidenced by its involvement in initiatives supporting RISC-V International.

Eliyan's team has created enterprise and server class semiconductors. The company consists of a team of PHY designers and chip architects who have produced generations of semiconductor products and have a profound understanding of systems. Eliyan's Interconnect technology, developed using ordinary organic chip packaging, gives the same performance as advanced packaging technology, allowing The Ultimate Chiplet Systems to be realized at a fraction of the total cost of ownership. You get more cores, computing performance, test coverage, and yield without the drawbacks of sophisticated packaging.

Movellus develops process portable and ultra-low power analog IPs. Its roducts include PLLs, DLLs, and LDOs, that are optimized per design.

Today’s explosive data growth and rapidly expanding application workloads threaten to overwhelm datacenters, preventing organizations from harnessing AI/ML, 5G, big data analytics, IoT, and other modern applications they need to achieve their business goals. Accelerating these data-hungry, storage-intensive applications requires more application speed, storage capacity, data efficiency, and scalability than most organizations can afford, no matter how many CPUs and SSDs they add, or how much they expand their datacenter footprint. Just as GPUs overcome processing inefficiencies to accelerate artificial intelligence and advanced analytics performance, the breakthrough Pliops Extreme Data Processor (XDP) overcomes storage inefficiencies to massively accelerate performance and dramatically reduce overall infrastructure costs for databases, analytics, and machine learning. Delivered on an easy-to-deploy, low-profile PCIe card, the Pliops XDP radically simplifies the way data is processed and flash storage is managed. The result is an exponential increase in performance, reliability, capacity, and efficiency— multiplying the effectiveness of organizations’ infrastructure investments.

NeuReality is a semiconductor start-up that designs AI-as-a-service infrastructure. It addresses the challenges posed by the growing deployment of AI in the cloud, edge, and fog systems.

D-Matrix is a computing platform that targets artificial intelligence inferencing workloads in the datacenter. Its mission is to build and deploy the world’s first brain-inspired AI compute engine for a broad class of inferencing workloads in the multi-billion dollar cloud and infrastructure edge markets.

Elastics.cloud, Inc. is a Smart Interconnect technology company focused on enabling efficient and performant architectures to create flexible, scalable, low latency composable systems. The company provides silicon, hardware, and software which leverages the Compute Express Link (CXL) interconnect standard to provide high-performance connectivity to a broad ecosystem of components.

Headquartered in Carlsbad, California, GigaIO democratizes AI and HPC architectures by delivering the elasticity of the cloud at a fraction of the TCO (Total Cost of Ownership). With its universal dynamic infrastructure fabric, FabreX™, and its innovative open architecture using industry-standard PCI Express/soon CXL technology, GigaIO breaks the constraints of the server box, liberating resources to shorten time to results. Data centers can scale up or scale out the performance of their systems, enabling their existing investment to flex as workloads and business change over time. For more information, contact info@gigaio.com or visit www.gigaio.com.

FMC is the provider of extremely efficient FeFET memory solutions for non-volatile storage. FeFET is ultra-low-power, high-performance, very dense, and extremely temperature stable. It will scale to transistor sizes ranging from 28nm to 5nm and below. The FMC team has been supported by “EXIST Forschungstransfer”, a program of the Federal Ministry for Economic Affairs and Energy. The company is based in Dresden, Germany.FMC commercializes a disruptive material innovation that will solve this problem for current and future technology nodes, i.e. eNVM based on ferroelectric hafnium oxide (FE-HfO2). The unexpected physical effect of ferroelectricity in HfO2 allows for the transformation of classical high-k metal-gate (HKMG) transistors into nonvolatile ferroelectric field effect transistors (FeFET). In this way, MCUs can easily be scaled from e.g. 65 nm down to 28 nm and beyond enabling tremendous advantages for the overall system: SoC cost reduction of around 80%, per bit write energy reduction by a factor of 1000 and an overall performance gain of the system of around 70% due to the transition to advanced process nodes. Due to the close relation of FeFET and CMOS baseline, there is no roadblock for FMC’s technology to be applied to cutting-edge technology nodes like 22 nm FDSOI, 1X nm FinFET and beyond.

Gauss Labs aims to revolutionize manufacturing by building industrial AI systems beyond human capabilities.

SiFive designs chips for the semiconductor industry. It offers the system on a chip for the microcontroller, embedded, IoT, and wearable applications; and toolset to provide debug capabilities. SiFive brings the power of open source and software automation to the semiconductor industry, making it possible to develop new hardware faster and more affordably than ever before. With the platform for designing, testing, and building RISC‑V-based core IP and chips, they’re accelerating the pace of innovation for businesses large and small.

MemVerge is delivering the world’s first Memory-Converged Infrastructure (MCI) system, built with proprietary Distributed Memory Objects (DMO) technology. DMO technology provides a logical convergence layer that harnesses Intel’s new Optane DC persistent memory to let data-centric workloads run flawlessly at memory speed.

Inpria Corporation is a pioneer in extending semiconductor lithography with inorganic photoresists and other thin-films for nanoscale patterning. Inpria’s patented photo-condensed molecular oxides offer customers the unique ability to deposit – directly from solution – atomically smooth, intrinsically dense, and directly photopatternable metal oxide films, enabling advanced performance with simplified processing.

Horizon Robotics is a leading provider of ADAS and AD solutions for passenger vehicles with proprietary software and hardware.  Our solutions combine cutting-edge algorithms, purpose-built software and state-of-the-art processing hardware, providing the core technologies for assisted and autonomous driving that enhance the safety and experience of drivers and passengers.

Zettastone is engaged in solid-state storage technology based on flash memory chips. They provide solutions for PC manufacturers, financial and other industrial control fields, data centers, industrial automation, transportation, medical treatment, and telecommunications.

TidalScale has developed a scale-up compute and resource architecture. It specializes in the fields of analytics, cloud infrastructure, information technology, software and data mining.  It was founded in 2013 and headquartered in Campbell, California.

Starblaze is a company that manufactures consumer and enterprise SSD master chips. The company's independently developed high-performance and low-power NVMe SSD master controller has been mass-produced and shipped, as has the firmware solution, which has supported many well-known Brand manufacturers who have launched high-performance NVMe SSDs. They also provide four different sets of solutions: STAR1200C, a high-performance, low-power consumption NVMe solution; OC SSD, a data center solution based on a customized Open-Channel interface; STAR1200E, a low-power, high-performance NVMe enterprise-level solution; and DeepSSD products based on AI storage master control, enabling KV storage applications in the monitoring market.

Reno Sub-Systems, to our longtime semiconductor friends, colleagues and customers around the world. The founding of Reno has been a very gratifying challenge which has brought together many of the best and brightest minds in the field of Semiconductor Process Equipment Instrumentation. At a time when industry consolidation and the tightening of venture capital poses a serious threat to new business formation, the founders of Reno have demonstrated the incredible power of inventing world-class technologies to enable next generation processes. With three brand new and highly disruptive technologies, we are creating a paradigm shift in the industry. At Reno, we believe there is unlimited potential for new discoveries and innovative thinking. As we move forward into the 450mm era, Reno stands at the summit of semiconductor process control equipment technology, ready to set the standards for product performance, worldwide customer support and customer satisfaction.

Exnodes detects 20 nanometer defects at 120 wafers per hour to maximize yield of computer chips. To license Exnodes technology and patents for your next generation product, please write to us at onboard@exnodes.com Using an 8.9 nanometer sensitivity breakthrough with visible light, computational parallel inspection (CPI) enables unpatterned wafer inspection without the baggage of DUV. While standard methods rely on brute force increasing of scattered signal levels with shorter wavelengths, CPI rethinks nanoscale defect detection with information theory and maximizes sensitivity by leveraging the maturity and gentleness of visible light. Exnodes is made possible by world’s leading semiconductor companies and by inventors, builders and nurturers determined to make the nanoscale accessible for all of us.

NetSpeed Systems provides scalable, coherent, Network-on-Chip (NoC) IPs to System-on-chip (SoC) designers for a wide range of markets from mobile to high-performance computing and networking. NetSpeed's NoC IPs deliver significant time-to-market advantages through system level approach, high level of user-driven automation and state-of-the-art algorithms.  NetSpeed Systems was founded in 2011 and is led by seasoned executives from the semiconductor and networking industries. The company is funded by top-tier investors from the silicon valley.

Stratio, Inc. was founded in 2013 with the mission to bring infrared sensing to the mass market. We created the world's first germanium-based shortwave infrared (SWIR) sensor, enabling portable, affordable and versatile infrared device creation. Today, we design, develop and manufacture portable spectrometers (LinkSquare®) and infrared cameras (BeyonSense®). Our partners and customers use Stratio's devices worldwide to increase process efficiency, fight counterfeiting, ensure food safety and promote precision agriculture. Continuing our efforts to innovate infrared sensing, Stratio is committed to develop Hyper-Spectral Imaging (HSI) solutions in the near future.

SK Hynix is the global leader in producing semiconductor, such as DRAM and NAND flash and System IC including CMOS Image Sensors. Since pilot production of Korea’s first 16Kb SRAM in 1984, SK Hynix consistently led the industry with smaller, faster and lower power semiconductor.

<div> SK hynix Inc. (KRX: 000660) is the world's second-largest memory chip manufacturer and a leading supplier of semiconductors. SK hynix produces dynamic random access memory (DRAM) chips, NAND flash memory chips, and CMOS image sensors (CIS) for a wide range of global customers. The company has four production sites in South Korea and China, sales subsidiaries in ten countries, and research and development centers in Italy, the US, Taiwan, and Belarus. SK hynix was founded in 1983 and is headquartered in Icheon, South Korea. SK hynix's main products and services include: <ul> <li> DRAM chips for servers, PCs, and mobile devices </li> </ul> <ul> <li> NAND flash memory for solid-state drives (SSDs) and mobile storage </li> </ul> <ul> <li> High Bandwidth Memory (HBM) for artificial intelligence and high-performance computing </li> </ul> <ul> <li> CMOS image sensors for mobile devices and automobiles </li> </ul> <ul> <li> Memory solutions for data centers and enterprise storage systems </li> </ul> SK hynix has been making significant investments to expand its production capacity and develop next-generation memory technologies. In April 2024, the company announced plans to invest USD 3.87 billion to build an advanced packaging facility for AI products in Indiana, US. SK hynix is also constructing the Yongin Semiconductor Cluster in South Korea, with an investment of KRW 120 trillion (approximately USD 90 billion) to build four state-of-the-art fabs. The company has been benefiting from the growing demand for AI-related memory chips. In Q4 2023, SK hynix reported its first quarterly operating profit in a year, with revenues of KRW 11.31 trillion (USD 8.6 billion) and an operating profit of KRW 346 billion (USD 263 million). The company expects strong growth in high-bandwidth memory (HBM) chips used in AI applications, with HBM revenue projected to grow by 300% year-on-year in 2024. </div>

SK Hynix to receives $958 mn subsidy from US

Platforms and frameworks that help developers build and train ML models.

Platforms that deploy and host the trained and tested models in production and integrate them into applications through APIs.

Platforms that generate artificial data that mimics the characteristics of real data, which is then used to train ML models at scale.

Platforms that help monitor and manage ML models in training and deployment to track performance, conduct variance analysis, and optimize performance.

Makers of hardware such as CPUs, GPUs, and other application-specific hardware required to train and run ML workloads.

Integrated solutions that handle the end-to-end processes of developing, training, deploying, and monitoring ML models.

Data exchange providers and platforms that offer data labeling services via manual and/or software methods, which are used as input to train ML models.

Companies that develop low-power, high-performance semiconductors for sensors, wearables, and other connected devices.

Companies that develop semiconductors that can support high-bandwidth and low-latency requirements of 5G networks and other wireless communication technologies.

Companies that develop semiconductors that focus on automotive applications such as advanced driver assistance systems (ADAS) and autonomous driving as well as in-car entertainment and communication systems.

Companies that develop semiconductors to improve the performance of data center infrastructure.

Companies developing semiconductors for other emerging tech applications, including quantum and edge computing.

SK Hynix, a South Korean chipmaker, has introduced a new device called AiMX, which improves AI system performance. AiMX is a GenAI accelerator card that uses GDDR6 chips to achieve faster data processing and lower power consumption than GPU-based systems. Benefits of AiMX include 10x faster data processing and greater energy efficiency.

SK Hynix

Featured Industries

About EDGE

Contact Us

Resources

Area of Focus

Technology Research Made Effortless