The artificial intelligence (AI) chips business is hot — red hot, by most accounts. Intel, Google, AMD, Arm, and others are vying for a market some analysts forecast will be worth $91 billion by 2025, and they’re not the only ones. Four-year-old San Francisco startup Esperanto Technologies has a stake in the race, too, and it’s making small but meaningful steps toward challenging the sector’s incumbents.
Esperanto this week announced a $58 million series B funding round led by “numerous” undisclosed venture and strategic capital investors, bringing its total haul to date to $63 million. CEO Dave Ditzel said the cash infusion will help to accelerate development of its first-generation chip lineup.
“Despite still operating largely in stealth mode, we appreciate this strong show of support from strategic and VC investors who had confidential briefings about our plans and believe we have a compelling solution for accelerating ML applications,” he said. “Esperanto has assembled one of the most experienced VLSI product engineering teams in the ML industry, and we believe that will be a differentiating factor as we drive toward our … products.”
The startup — which counts Western Digital as one of its previous investment partners — aims to develop energy efficient, high-performance compute solutions based on RISC-V, an open source and royalty-free instruction set architecture (ISA). RISC-V isn’t the first open computing architecture, but it’s designed to be useful in a wide range of devices and has a substantial body of supporting organizations that includes Google, Hewlett Packard Enterprise, IBM, Qualcomm, Oracle, Nvidia, and others.
June 5th: The AI Audit in NYC
Join us next week in NYC to engage with top executive leaders, delving into strategies for auditing AI models to ensure fairness, optimal performance, and ethical compliance across diverse organizations. Secure your attendance for this exclusive invite-only event.
With its forthcoming 64-bit 7-nanometer processor, Esperanto says it’ll leverage standards such as the Open Compute Platform (OCP), Facebook’s Pytorch framework and Glow compiler, and the Open Neural Network Exchange (ONNX) to accelerate AI and machine learning workflows. The aforementioned chip, the design of which will be licensable, packs more than a thousand of Esperanto’s ET-Minion RISC-V cores — cores designed to deliver the best teraflop performance per watt efficiency, according to the company — on a single die, with a distributed memory architecture Esperanto claims “improves processing utilization” and “relieves memory bandwidth bottlenecks.”
ET-Maxion — Esperanto’s other core product offering — implements features such as quad-issue out-of-order execution, branch prediction, and prefetching algorithms, and can run high-level functions such managing on-chip data movement and scheduling. It works as either an accelerator or bootable core, serving in the latter configuration as an interface to system software layers.
Esperanto isn’t a fly-by-night operation. Ditzel, who previously founded x86 chip designer Transmeta, is the former vice president and chief architect for Intel’s hybrid parallel computing division and chief technology officer for Sun Microsystems’ SPARC technology business. David Glasco, Esperanto’s vice president of engineering, was previously the architecture and design lead for Tesla’s Autopilot system-on-chip hardware. And chief architect Roger Espasa currently co-chairs the RISC-V Foundation vector extensions task group and is active on the RISC-V technical committee.
Additionally, Esperanto has more than 100 employees on its payroll, including AI experts, processor architects, chip designers, software developers, and system engineers from Intel, DEC, MIPS, Sony Interactive Entertainment, and QED.
“Next-generation applications in machine learning, AI, and real-time analytics require the highest levels of performance and optimization for these advanced workloads,” Martin Fink, chief technology officer of Western Digital, said in a statement. “The RISC-V platform, and Esperanto solutions, free developers to innovate and optimize for special-purpose computing.”
