South Korean AI chipmaker Rebellions made the headlines in July 2024, by raising USD 15 million in a Series B extension. The funding round was led by Wa’ed Ventures, a venture capital firm owned by Saudi Aramco.
The funds will be used to expand the start-up’s business in Saudi Arabia, along with accelerating its plans for AI chip development. Rebellions also aims to establish a subsidiary in the Gulf nation, apart from launching new business operations in the region.
Founded in 2020, Rebellions specialises in AI inference accelerators that offer energy efficiency and low-latency performance. Within three years of its formation, the company has introduced two AI chips, with its third ready to be rolled out in the second half of 2024.
Rebellions completed its USD 124 million Series B funding this January, and with this extension, has now secured over USD 225 million in total funding since inception.
In today’s episode of the “Start-up of the Week,” International Finance will talk in detail about the venture.
Knowing Rebellions
The tech start-up is driven by generative AI and situated at the heart of the “Korean Silicon Ecosystem.” In 2023, the start-up made a thunderous entry into the AI chip market, by unveiling its maiden product, which was powered by a cutting-edge 5nm technology. Since then, the venture has made its presence felt, from Seoul to New York, San Jose, Tokyo, Singapore, and beyond, as it expanded its footprint on a war footing.
As per Park Sung-hyun, co-founder and CEO of Rebellions, what the start-up brings on table is its focus on developing more energy-efficient chips solely dedicated to AI services. And investors have been receptive towards the idea, as reflected in the successful funding rounds since the venture’s inception.
Rebellions was founded by a group of chip experts who worked at global tech companies like Samsung Electronics, SK Hynix, Intel and IBM. Park himself built his career at Intel Labs, Samsung Research America, SpaceX and Morgan Stanley.
The start-up drew immediate attention from the nation’s fabless industry when it released its first chip for AI-based stock trading, called ION, in November 2021. The chip is now widely adopted by several Wall Street investment firms, including Soft Bank-backed Qraft Technologies.
Another breakthrough came in 2023 as Rebellions launched another AI chip for data centres, named ATOM, to compete against American chip giant Nvidia, the dominant market leader with an almost 90% market share. It is also the first Korea-made large language model chip.
The company’s competitive edge has been its focus on developing a neural processing unit, a more power-efficient machine learning accelerator, than a graphic processing unit that is more widely produced by its bigger rivals, including Nvidia.
In June 2022, the South Korean government unveiled its plans to foster a domestic industry, investing 1.2 trillion won over the next five years for research and development in a bid to increase the market share of local AI chips in domestic data centres to 80% by 2030. Rebellions now look to take centre stage of this transformation effort.
In October 2023, Rebellions announced a strategic partnership with Samsung Electronics to co-develop their next-generation AI chip, Rebel. The chip will be produced using Samsung’s cutting-edge 4-nanometre fabrication process and will feature Samsung Electronics’ HBM3E memory technology.
Entering The Rebellions Ecosystem
Rebellions’ small yet powerful chips have the potential to be the game-changers across diverse fields, from finance to hyperscale data centres, as AI dawns further. Talking about Rebellions’ next-generation AI chip REBEL, will be a chiplet-based AI accelerator, specially engineered to enhance Large Language Models (LLMs) and Multi-Modal Models (MMMs).
Samsung and Rebellions aim to complete the development of Rebel by the 2024 end and start mass production in 2025, with the next-generation AI chip targeting the generative AI market running large language models (LLMs) and hyperscalers.
Since Samsung, South Korea’s largest memory chip maker, has been working on its own generative AI model, Samsung Gauss, the partnership with Rebellions and the development of REBEL will be equally helpful for the tech giant’s market prospects.
Next, we have ATOM, which is an inference AI accelerator for data centres. ATOM is basically a fast and power-efficient System-on-Chip for AI inference with remarkably low latency, conceived for deployment in data centres or cloud service providers.
The product is a versatile yet energy-efficient AI system-on-chip, delivering top performances across various types of AI tasks like computer vision, natural language processing and recommendation models. Utilising the silicon-proven neural core ION as the compute granule that scales up with perfect linearity, ATOM has become the optimal chip for large-scale inference required in edge computing and data centres.
ATOM is poised to be a pivotal enabler for server-level AI services, optimising the AI-as-a-Service (AIaaS) stack. As a multi-core System-on-Chip (SoC), the chip has been built upon dataflow architecture to deliver superior inference performance.
Using Samsung’s advanced EUV process node, ATOM has been delivering improved performance, reduced power consumption and enhanced energy efficiency. Geared with PCIe Gen5, GDDR6 and high-speed I/O, ATOM has transformed itself as the optimal AI accelerator to serve different markets, spanning from edge computers to data centres.
In fact, ATOM has shown impressive results across both vision and language models, outperforming Qualcomm and NVIDIA, respectively. The chip is built on the small yet powerful silicon-proven AI core ION, which, with its Coarse-Grained Reconfigurable Architecture (CGRA), is flexible, programmable, and scalable. This design allows ATOM to handle networks with varying depths and complexities, from small applications to hyperscale services, all while maximising energy efficiency.
The product is a multi-core “System-on-Chip,” consolidating all essential components onto a unified substrate. The communication overhead within the chip is masterfully handled via on-chip data dependency handling. It can also accelerate different types of neural networks, including Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM), Bidirectional Encoder Representations from Transformers (BERT) and transformers-based models.
ATOM can be partitioned to support up to 16 separate jobs simultaneously, with each job isolated at both the hardware and software levels. The chip’s “Multi-Instance Feature” allows higher resource usage and flexibility, optimising utilisation and performance.
Meet Rebellions’ ION
The chip we are going to talk about now, in the start-up’s words, will revolutionise the high-frequency trading market.
Being a “Scalable AI Compute Core” product, ION is a “Compute Granule with Maximum Flexibility and Efficiency.” It provides flexible inference capabilities with low power consumption, a small footprint, and high performance for edge computing systems.
Featuring a customised instruction set architecture (ISA) designed for over 1,000 multiply-and-accumulate (MAC) units, ION delivers inference acceleration with an exceptionally high utilisation rate compared to other AI accelerators. ION’s versatility, compact size, and low power requirements make it the optimal choice for edge deployments.
Powered by ION’s hardware and software full-stack implementation, LightTrader, the Field Programmable Gate Array (FPGA) integrating multiple chips, has been deployed in high-frequency trading (HFT) solutions at Wall Street-based investment banks. LightTrader achieves up to 60x AI-enabled HFT performance, instantly upgrading any HFT solution to a more intelligent trading system.
ION achieves up to 10x more performance-per-watt figure even in comparison against the state-of-the-art mobile Neural Processing Units (NPUs). This makes the product ideal for various applications including mobile companion chips, smart cities, robots and retail.
When it comes to its practical applications in the financial market, ION brings its game-changing “Innovative Trading Solution Architecture” into the play.
“The first-generation Compute Core ION supports FP16-based accurate stock prediction, utilising the single batch inference and predictive execution pipelines. ION also supports BFloat16 and low precision integer operations such as INT8/4/2,” Rebellions noted.
Promising Future Ahead
Apart from Samsung, Rebellions has also been in a partnership with Synopsys, another global tech venture. Synopsys, which is dedicated to empowering its customers with cutting-edge tools and technologies that drive innovation and excellence, has found a worthy customer in the form of a South Korean start-up.
The collaboration centres on Synopsys’ ZeBu emulation, Virtualizer, and VCS solutions, vital tools in the chip design and software development process. ZeBu together with Virtualizer enables Rebellions to emulate and verify the functionality of its AI accelerators executing AI workloads in a virtual environment, significantly speeding up the validation process.
Rebellions is facing a dual challenge: designing innovative AI accelerators and then validating their functionality and performance at scale. The ZeBu platform plays a pivotal role here by allowing Rebellions’ engineers to emulate complex workloads while fostering a mature full-stack software environment from the early stages. This approach not only reduces time-to-market but also enhances the reliability of the final product.
Rebellions, which is targeting expansion into key international markets such as Japan, the United States, and the Middle East, has been participating in global AI summits and conferences, in order to establish a global footprint and foster new partnerships. Through the partnership with Synopsys, Rebellions is enhancing the performance and reliability of its AI accelerators, apart from accelerating their path to market readiness. Rebellions partnerships with Samsung and Synopsys will not only help the start-up to navigate the complexities of AI chip design and meet the evolving demands of global IT markets, but the venture may also become a tech sector disruptor in the coming days.