Nguồn vốn khởi nghiệp: Quý 2 năm 2024
AI drew more investors to the chip industry in Q2. Four AI-focused chip startups receiving rounds of more than $100 million, targeting data center ASICs for transformers, highly flexible platforms for the embedded edge, dataflow processors, and mixed-signal neuromorphic chips. In-memory computing also helped boost AI, with three companies either incorporating it into their chips or providing specialized IP.
Ways to connect chiplets, chips, and server racks was another area of activity, with Germany pumping more than $200 million into a graphene-based co-integrated optics system for massively parallel connectivity. Other optical interconnect approaches drew attention, as well, along with a way to make ultra-high density interconnects using liquid metal ink.
This report covers 91 startups that collectively raised $2.6 billion in the second quarter of 2024.
Chips
MetisX raised $44.0M in Series A funding from SV Investment, STIC Ventures, LB Investment, and the Industrial Bank of Korea, and existing investors including Mirae Asset Venture Investment, Mirae Asset Capital, IMM Investment, SBI Investment, Tony Investment, and Wonik Investment Partners. MetisX is a fabless company developing a memory-centric data domain specific architecture based on CXL to overcome performance bottlenecks associated with frequent memory accesses. The startup says its intelligent memory architecture accelerates large-scale data processing applications, such as vector databases, big data analytics, AI data preprocessing, and DNA analysis. It plans to manufacture its first chip by early next year. Founded in 2022, it is based in Seongnam, South Korea.
Flow Computing emerged from stealth with €4.0M (~$4.3M) in pre-seed funding led by Butterfly Ventures and joined by FOV Ventures, Sarsia, Stephen Industries, Superhero Capital, and Business Finland. Flow Computing offers what it calls a Parallel Processing Unit (PPU) that can be integrated on-die into any CPU design architecture, instruction set, or process geometry with minimal CPU modifications. The startup claims the PPU IP can provide a 100X performance boost to CPUs through hiding the latency of memory references by executing other threads while accessing the memory, reducing the number of times threads need to be synchronized and overlapping them with the execution, and organizing functional units as a chain where a unit can use the results of its predecessors as operands. It is customizable to specific requirements, including the number of PPU cores, the variety and number of functional units, and the size of on-chip memory resources. A spinout from the VTT Technical Research Center of Finland founded in 2024, it is based in Helsinki, Finland.
Morphing Machines drew $2.8M in seed funding led by Speciale Invest, joined by IvyCap Ventures, Golden Sparrow, Navam Capital, CIIE Initiatives, and DeVC. Morphing Machines develops a many-core, massively parallel SoC platform that can scale from 16-cores to 4K-cores and is runtime reconfigurable, enabling domain specific architectures for mixed critical application tasks to be instantiated on demand upon an event. It targets applications such as avionics, automotive, and 5G/6G telecom. Founded in 2006, it is based in Bengaluru, India.
AI hardware
Rivos drew over $250.0M in Series A-3 funding led by Matrix Capital Management, joined by new investors including Intel Capital, MediaTek, Cambium Capital, CIDC Consultants, Capital TEN, and Hotung Venture Group, along with existing investors Walden Catalyst, Dell Technologies Capital, Koch Disruptive Technologies, and VentureTech Alliance. Rivos designs software-defined hardware solutions for data center workload acceleration, particularly data analytics and generative AI. Its chips combine high performance server-class RISC-V CPUs and a general-purpose GPU (GPGPU) optimized for LLMs and data analytics that share a uniform memory across DDR DRAM and HBM. It also offers a firmware-to-application open software stack. Funds will be used to tape out its first silicon product and expand manufacturing operations, platform hardware, software engineering, and support functions. Founded in 2021, it is based in Santa Clara, California, USA.
Etched raised $120.0M in Series A funding led by Primary Venture Partners and Positive Sum Ventures, with participation from Two Sigma Ventures, Skybox Datacenters, Hummingbird Ventures, Oceans Ventures, Fundomo, Velvet Sea Ventures, Fontinalis Partners, Galaxy Digital, Earthshot Ventures, Max Ventures, Lightscape Partners, and individual investors. Etched is building an algorithm-specific ASIC targeted solely for transformer model inference. Transformers are currently the dominant architecture used for generative AI. The startup says that by only running one algorithm, the vast majority of control flow logic can be removed, allowing it to have many more math blocks, which results in over 90% FLOPS utilization. Etched claims that one server containing eight of its Sohu chips can run over 500,000 Llama 70B tokens per second. Founded in 2022, it is based in Cupertino, California, USA.
Hailo extended its Series C round with $120.0M in new funding led by Delek Motors, DCLBA, Vasuki, OurCrowd, Talcar, Comasco, Automotive Equipment Group, Poalim Equity, and individual investors. Hailo makes AI processors for edge devices. It has developed a domain-specific structure-defined dataflow processor architecture that allows software to allocate compute, memory, and control blocks as needed for different neural network layers. The startup recently debuted a new AI accelerator specifically designed to process large language models (LLMs) at low power consumption for generative AI in personal computers and automotive infotainment systems. It also offers general AI processors and a series of AI vision processors for smart cameras, with products available as SoCs and modules, alongside a full software stack. Founded in 2017, it is based in Tel Aviv, Israel, and has raised over $340M to date.
Blaize received $106.0M in funding from existing investors including Bess Ventures, Franklin Templeton, DENSO, Mercedes Benz, and Temasek, and new investors Rizvi Traverse, Ava Investors, and BurTech LP. Blaize provides a full-stack programmable graph streaming processor architecture suite and a low-code/no-code software platform for edge AI in industries such as automotive, mobility, retail, security, industrial automation, and healthcare. The company’s graph streaming processors are the basis for a range of products including system-on-modules, M.2 cards, PCIe accelerator cards, and an embedded kit. Blaize is in the process of going public on Nasdaq through a merger with SPAC BurTech Acquisition Corp. Founded in 2010, it is based in El Dorado Hills, California, USA.
DeepX received $80.5M in Series C funding led by SkyLake Equity Partners with participation from BNW Investments, AJU IB Investment, and TimeFolio Asset Management. DeepX is developing on-device NPU IP and AI SoCs tailored for applications including physical security, robotics, and mobility. The startup claims that by compressing FP32 into Int8 without a reduction in accuracy and by reducing DRAM accesses, its NPU SoC can achieve more than 10 TOPS/W computational efficiency with power consumption of less than 3 watts. It also offers NPU design services, AI modules, and a software design kit. Funds will be used for mass production of its first product line and development and launch of its next-generation LLM on-device solutions. Founded in 2018, it is based in Seongnam, South Korea.
SiMa.ai drew $70.0M in new funding led by Maverick Capital, with participation from Point72 and Jericho Capital, as well as existing investors Amplify Partners, Dell Technologies Capital, Fidelity Management & Research Company, Lip-Bu Tan, and others. SiMa.ai builds hardware and software for machine learning workloads at low power. The startup claims its platform can adjust to any framework, network, model, sensor, or modality. It currently offers a computer vision inference chip for the embedded edge in sectors such as manufacturing, retail, aerospace, defense, agriculture, and healthcare. Its software-centric MLSoC platform aims to be a simple way to deploy and scale edge ML and includes computer vision processors, video encoder/decoder, ML accelerator, and application processors. Its software enables static scheduling and double buffering that allows for proactive data prefetch in a layered approach ahead of the compute time, which SiMa.ai claims enables it to run models of any size. Available as an SoC and PCIe-based production boards, the platform can be used either as a standalone edge-based system controller or as a ML offload accelerator for processors and ASICs. The startup is currently working on its second-generation chip for multimodal generative AI at the edge, targeted for early 2025. Founded in 2018 and based in San Jose, California, USA, it has raised $270M to date.
Axelera AI raised $68.0M in a Series B round that included Invest-NL, the European Innovation Council Fund, and Samsung Catalyst Fund, along with existing investors Verve Ventures, Innovation Industries, Fractionelera, and CDP Venture Capital SGR. Axelera AI provides an edge AI acceleration platform for generative AI and computer vision inference that utilizes SRAM-based digital in-memory computing and a RISC-V controlled dataflow architecture to minimize data movement between memory and compute elements. It also offers quantization techniques and mapping tools that reduce AI computational load and increase energy efficiency while minimizing accuracy loss for a wide range of networks. The startup anticipates that its AI processing unit will be in full production in the second half of 2024. Funds will be used to expand into new markets, including automotive and high-performance computing. Founded in 2021 and based in Eindhoven, the Netherlands, it has raised $120M to date.
Expedera raised $20.0M in Series B funding led by indie Semiconductor. Expedera provides high-performance, energy-efficient Neural Processing Unit (NPU) IP for on-device AI. Targeting applications such as automotive, AR/VR/MR, and smartphones, its NPU IP scales to 128 TOPS with a single engine and requires minimal external memory bandwidth. It uses a packet-based architecture that subdivides each layer into self-contained executable fragments that can be scheduled independently, enabling parallel execution across multiple layers, better resource utilization, and deterministic performance. The platform includes a TVM-based software stack that allows the importing of trained networks, provides various quantization options, automatic completion, compilation, estimator, and profiling tools. Funds will be used to advance its IP and services and to expand its global engineering footprint. Founded in 2018, it is based in Santa Clara, California, USA.
Innatera added $5.0M to its Series A round in an extension led by Innavest and Invest-NL, bringing the round to $21.0M. Innatera develops neuromorphic processors based on an analog-mixed signal computing architecture that mimic the brain’s mechanisms for processing sensory data. Innatera’s processors use spiking neural networks with ultra-low power consumption and short response latency for always-on pattern recognition capabilities in applications at the sensor-edge. Funds will be used to accelerate mass production. Founded in 2018 as a spin off from the Delft University of Technology, it is based in Rijswijk, the Netherlands.
RaiderChip drew €1.0M (~$1.1M) in seed financing. RaiderChip is developing a hardware IP core for accelerating generative AI inference on FPGA platforms. It says its approach enables small to large language models to run locally on low-cost standalone or embedded devices by employing massive floating-point parallelism and optimized memory bandwidth utilization. Funds will be used for marketing. Founded as a spin-off from Visengi in 2024, it is based in Solares, Spain.
EDA
Itda Semiconductor drew KRW 3,000.0M (~$2.2M) in pre-Series A financing from We Ventures and L&S Venture Capital. Itda Semiconductor provides a tool for no-code design and optimization of power and clock systems in complex SoCs. Founded in 2022, it is based in Hwaseong, South Korea.
Manufacturing & equipment
Eyco received €16.0M (~$17.4M) in financing from Bpifrance, Région Sud Investissement, CappCreation, and others. Eyco manufactures flexible circuits on ultra-thin film substrates that are only a few microns thick. It has developed on-reel processing equipment with capabilities such as multi-layer and multi-material co-lamination, mechanical cutting, offset printing, etching, metallic surface treatment, and electrical sensitization of insulating materials and micro-vias. It focuses on devices for the medical, payments, security, and telecommunications industries. Funds will be used for industrialization. Founded in 2020, it is based in Trets, France.
Axus Technology drew $12.5M in capital funding from IntrinSiC Investment. Axus Technology makes chemical mechanical planarization (CMP) equipment for semiconductor and compound semiconductor wafer polishing, thinning, and cleaning. Its surface processing solutions cover semiconductors including silicon carbide, MEMS device processing, device packaging, and TSV/3D integration. Funds will be used to pursue and fulfill high-volume orders. Founded in 2002, it is based in Chandler, Arizona, USA.
Test, measurement & inspection
Liquid Instruments drew $12.0M in funding led by Breakthrough Victoria, joined by Lockheed Martin Ventures, Acorn Capital, and Powerhouse Ventures. Liquid Instruments makes reconfigurable test and measurement devices that use FPGAs to combine multiple instruments into one. Its platform includes oscilloscope, programmable power supply, PID controller, digital logic analyzer, arbitrary waveform generator, data logger, spectrum analyzer, and more. The company’s offerings range from portable devices aimed at engineering education to a more advanced device aimed at research and professional engineering that allows users to run and connect multiple instruments at the same time for a customized test system. It targets a wide range of applications from a range of applications, from optics and photonics to aerospace and defense. Funds will be used to expand its manufacturing operations, establish a new office, and scale its global operations. Founded in 2014, it is based in Canberra, Australia.
Materials
Thintronics drew $23.0M in Series A funding led by Maverick Capital and Translink Capital. Thintronics makes high-performance interconnect insulators for AI data center, networking, and RF/millimeter-wave (mmW) applications. The startup says its ultra-thin dielectric layers can be used in the die, substrate, and PCB across compute, switches, and networks for interconnects up to 224 Gbps with reduced channel loss and 50% thickness reduction. Founded in 2019, it is based in Berkeley, California, USA.
LQDX received $10.0M in equity and government funding. LQDX develops a liquid metal ink metallization chemistry suite that enables nanoscale layers of pure palladium, gold, or copper to be applied atom-by-atom to a range of substrate materials, including 3D surfaces, for advanced packaging interconnects in chips, chiplets, and heterogeneous integration. The company says its technology enables production of much denser circuits than conventional approaches and is suitable for ultra-high density interconnect substrates, complex multi-layer interposers, and advanced wafer-level fan-out techniques. Founded in 2007 as Averatek, it is based in Santa Clara, California, USA.
Memory & storage
Synthara raised $11.4M in equity financing and grants led by Vsquared Ventures, joined by OTB Ventures, Onsight Ventures, Deeptech Labs, and existing investors including High-Tech Gründerfonds, DeepIE Ventures, Excellis, and Zürcher Kantonal Bank. Synthara provides a drop-in replacement to SRAM memory with integrated computing capabilities. It requires no foundry waivers, is compatible with any CMOS process and ISA, and includes an SDK with optimized implementations of linear algebra subroutines, signal processing functions, and neural network layers. The compute-in-memory technology is targeted at embedded microcontrollers, which the startups claims can see a 100x boost to energy efficiency and latency by offloading up to 99% of model operations to memory, eliminating the need for dedicated AI accelerators in wearable, robotics, and smart sensing applications. Founded in 2019 as a spin-off from the Institute of Neuroinformatics at ETH Zurich and the University of Zurich, it is based in Zurich, Switzerland.
Rain AI drew $8.1M in a Series A extension round from Epic Venture Partners. Rain AI offers digital in-memory compute IP that is tailored to on-device AI workloads requiring ultra-low latency and high energy efficiency. It supports both training and inference and uses quantization algorithms that are co-designed at the circuit level to maintain FP32 accuracy. The startup has also developed a proprietary interconnect between RISC-V and its D-IMC cores that enables AI developers to implement any operator and compile any model. Its first chip is expected to launch in 2025 and will target applications such as drones, VR, smartphones, robotics, and wearables. Founded in 2017, it is based in San Francisco, California, USA.
ZeroPoint Technologies raised €5.0M (~$5.5M) in Series A funding led by Matterwave Ventures, joined by Industrifonden, Climentum Capital, and Chalmers Ventures. ZeroPoint develops memory compression IP. The company says its technology can reduce the energy consumption of data centers by combining ultra-fast data compression with real-time data compaction and transparent memory management in an IP block that integrates with existing industry standard on-chip bus protocols. Targeted for both data center servers and smart devices, the technology can compress data across the entire memory hierarchy and is agnostic to data load, processor type, architectures, memory technologies and processing node. Funds will be used to scale sales, bring additional hardware-accelerated memory products to market, and hire. Founded in 2016 as a spin out from Chalmers University of Technology, it is based in Göteborg, Sweden.
RAAAM Memory Technologies drew $4.0M in seed funding from Serpentine Ventures, J-Ventures, HackCapital, Silicon Catalyst Angels, Claves Investments, and others. RAAAM has developed an on-chip memory technology it calls Gain-Cell Random Access Memory (GCRAM) that only requires 3 transistors to store a bit of data. The GCRAM bit-cell utilizes decoupled write and read ports, providing native two-ported operation. Proposed as a drop-in SRAM replacement with lower area and power consumption, it is manufacturable using a standard CMOS process. Founded in 2021, it is based in Petach Tikva, Israel.
Photonics & optics
Black Semiconductor received €254.4M (~$273.9M) between its Series A, led by Porsche Ventures and Project A, joined by East Hill Equity, Capnamic, TechVision Fonds, and NRW.BANK, and public funding under Europe’s IPCEI ME/CT program. Black Semiconductor has developed a co-integrated optics system based on graphene as translator to convert electronic signals into optical signals and vice versa. The startup says its approach enables massive parallel optical chip connectivity up to kilometers in distance. Funds will be used to drive its product development and 300mm wafer pilot production facility forward, with a long-term goal of mass production in 2031. Founded in 2019, it is based in Aachen, Germany.
Wave Photonics received £4.5M (~$5.8M) in seed funding led by the UK Innovation & Science Seed Fund and Cambridge Enterprise Ventures with participation from the Redstone QAI Quantum Fund, Kyra Ventures, Parkwalk Advisors, and Deep Tech Labs. Wave Photonics uses computational techniques to create tools and IP for integrated photonics design. Its approach uses a combination of simulation, statistical modeling, and optimization techniques. The startup offers a PDK along with a photonic chip packaging service for prototyping and R&D. Funds will be used to take its technology from a research manufacturing line to a commercial foundry, with a particular focus on solutions for applications such as quantum technologies and biosensing. Founded in 2021, it is based in Cambridge, UK.
NcodiN received €3.5M (~$3.8M) in funding from Elaia Partners, Earlybird Venture Capital, and OVNI Capital. NcodiN is developing optical interposer technology for chiplet integration. Through on-chip integration of ultra-small III-V semiconductor nanolasers and nanodetectors, the startup says its interposer enables ultra-high integration density of more than 10,000 components per cm2 while increasing data transfer speeds and efficiency. The startup is particularly focused on chiplet-based disaggregated processors and memory-centric computing architectures for high-performance computing and AI workloads. Funds will be used to accelerate the development of the integrated optical link prototype and for company expansion. Founded in 2023, it is based in Palaiseau, France.
Ranovus received a CAD 4.8M (~$3.5M) repayable investment from FedDev Ontario. Ranovus develops monolithic silicon photonic interconnects for AI/ML workloads, high-performance computing, and hyperscale data centers. The startup’s technology uses quantum dot multi-wavelength lasers (QD MWL) that can provide multiple wavelengths simultaneously from a single device, each of which can be selected and used as a light source for data transmission. It is capable of generating up to 96 wavelengths in the C-band. A silicon photonic micro-ring resonator architecture enables high-speed modulation at data rates of up to 100Gbs / lambda. These elements are combined with RF drivers, TIA, and control logic using a wafer scale laser attach packaging process. Funds will be used to scale commercial in-house production. Founded in 2012, it is based in Ottawa, Ontario, Canada.
Lightium was awarded a CHF 2.7M (~$3.0M) grant from InnoSuisse. Lightium has developed a thin-film lithium niobate (TFLN) process for manufacturing and packaging of photonic ICs. The startup, which intends to offer foundry services, says its TFLN platform has capabilities to support data transmission speeds exceeding 1.6 Tb/s. It initially targets designs for the telecommunication and data center sectors, but says the platform has applications in markets such as satellite communication, quantum computing, and lidar. Funds will be used to fine-tune its PDK and transfer its manufacturing process to a commercial CMOS fab to enable volume production. Lightium expects to open beta access to its TFLN platform by the end of 2024. Founded in 2023, it is based in Zürich, Switzerland.
Sensors
Bitsensing drew $25.0M in Series B financing from Korea Development Bank, HL Mando Corporation, Industrial Bank of Korea, Aju Capital, Life Asset Management, and SCL Investment. Bitsensing makes a range of radar systems, including 79GHz 4D imaging radar for automotive with a range coverage of more than 300 meters as well as a 24GHz radar with integrated AI perception for intelligent traffic management. It has expanded to the wellness market with a radar that monitors sleep quality, apnea events, limb movements, and also provides data analysis. Funds will be used to scale operations and for R&D. Founded in 2018, it is based in Seoul, South Korea.
Security
TXOne Networks raised $51.0M in Series B extension funding from existing investors TGVest Capital, Pegatron Group, CDIB Capital Group, CDIB-Innolux, and new investors Taiwania Capital and Applied Ventures ITIC Innovation Fund. TXOne Networks offers cybersecurity solutions to ensure the reliability and safety of industrial control systems (ICS) and operational technology (OT) environments through a zero-trust methodology. It provides both network-based and endpoint-based products that use a real-time, defense-in-depth approach to secure the OT network and mission-critical devices in a range of industries, including semiconductor facilities. Formed as a joint venture between Trend Micro and Moxa Technologies in 2019, it is based in Taipei, Taiwan.
PQShield received $37.0M in Series B financing led by Addition, with participation from Chevron Technology Ventures, Legal and General, Braavos Capital, and Oxford Science Enterprises. PQShield provides hardware components, sub-systems, and accelerators for post-quantum cryptography algorithms and side-channel analysis protection. It also offers software implementations of both post-quantum and classical cryptographic primitives. Founded in 2018, it is based in Oxford, UK.
Niobium Microsystems drew $5.5M in seed financing. Niobium makes SoCs and hardware accelerator cards for fully homomorphic encryption (FHE) that combine massively parallel computation circuitry with compilers optimized for memory efficiency at large scale, maximizing data re-use during computation. Funds will be used to develop commercial applications for FHE acceleration, for hiring, and for product optimization. Founded in 2021 as a spin out from Galois, it is based in Dayton, Ohio, USA.
Other technologies
Frore Systems raised $80.0M in Series C funding led by Fidelity Management & Research Company, joined by Prosperity7 and existing investors including Mayfield Fund, Clear Ventures, Addition, Qualcomm Ventures, MVP Ventures, Stepstone Group, and Alumni Ventures. Frore Systems makes solid-state active cooling modules that sit atop a chip to remove heat without fans. The piezoelectric MEMS device incorporates tiny membranes that vibrate at ultrasonic frequency to draw in a flow of air, which is then turned into high velocity pulsating jets of air that remove heat from the heat spreader. The startup claims the modules can remove more heat than fans while being audibly silent, vibration-free, and dustproof. It targets a range of systems, including notebooks and other consumer devices to network edge AI gateways and data centers. Founded in 2018, it is based in San Jose, California, USA, and has raised $196M to date.
Table
Funds & investors
imec.xpand launched a new €300.0M (~$320.9M) fund that will invest in semiconductor and nanotechnology startups with the potential to push semiconductor innovation beyond traditional applications and drive next-generation technologies.
Silicon Catalyst Ventures formed with the aim of investing $10.0M to $20.0M in its first series this year. It will back early-stage semiconductor startups primarily in North America, UK, EU, and Israel, covering AI, communications, photonics, MEMS, sensors, IP, materials, and life sciences.
PhotonVentures added €15.0M (~$16.0M) in its second fundraising round, giving it €75.0M ($79.8M) to invest in startups and scale-ups developing solutions with photonic chips.
Intel Ignite selected startups across the US, UK, Europe, and Israel to participate in its accelerator program for early-stage deep tech startups. Selected startups include those working on EDA, AI chips, photonic interconnects for advanced packages, memory compression IP, holographic data storage, and conformal imprint lithography.
China is investing CNY 344B (~$47.5B) into the third phase of its National Integrated Circuit Industry Investment Fund, also known as the Big Fund, to support its semiconductor sector and supply chain.
