We are excited to bring Transform 2022 back in-person July 19 and virtually July 20 - 28. Join AI and data leaders for insightful talks and exciting networking opportunities. Register today!
Today, Ceremorphic, a startup, is coming out of stealth mode with $50 million of series A funding along with the announcement of the QS 1 chip, which it claims can address multiple high-performance computing workloads from AI training to metaverse computing. Its Hierarchical Learning Processor (HLP) is built on TSMC’s 5nm node and the company claims to support PCIe 6.0 and CXL 3.0
Backed by over 100 patents, one of Ceremorphic’s key technologies is what it calls its ThreadArch technology for multi-threaded processing. Although no further details were provided, the HLP also contains several key IP blocks, including a custom machine learning processor and a floating-point unit, both of which operate at 2GHz. Besides these AI and HPC accelerators, the QS 1 further contains a multi-threaded RISC-V CPU and a M55 V1 GPU (which Ceremorphic claims is targeted for metaverse processing) that operate at 1GHz.
Ceremorphic’s claims of support for PCIe 6.0 and CXL 3.0 would make the QS 1 the first chip to support these standards. The PCIe 6.0 standard has just recently been finalized (with first silicon expected in 12 to 18 months), and CXL 3.0 hasn’t even been announced. The chip further supports open AI frameworks and support libraries.
QS 1 to support PCIe 6.0 and CXL 3.0
No further details or specifications such as transistor count, chip size, and performance targets were provided, so it’s hard to say how competitive it will be against solutions from existing chip vendors (such as Nvidia and Intel) as well as other start-ups (such as Cerebras, Graphcore, and Habana). However, Ceremorphic said its test chip would come out of TSMC’s fabs in March, which means the design must already have been taped out. After this milestone, it typically takes about another year to reach commercialization.
The applications that Ceremorphic seeks to address with its QS 1 chip include AI training, high-performance computing (HPC) for industries such as automotive, robotics and healthcare, and metaverse computing. The design targets were reliability, security, and energy consumption, although the only claim in that regard was a software error rate of one in ten thousand.
The startup was founded in April 2020 and employs 150 people, with plans to hire 100 more in 2022. R&D began more than five years ago – more than three years prior to Ceremorphic itself was founded by industry veteran Venkat Mattela, who’s also the founding CEO of Redpine Signals, a wireless connectivity company focused on low power. This company sold its wireless assets to Silicon Labs in March 2020 for $308 million, after which Mattela founded Ceremorphic with the computing assets that it retained.
“Having developed many innovations in multi-thread processing, low energy network circuits, analog computing, quantum resistant security microarchitecture, and new device architectures beyond CMOS, Ceremorphic is well on its way to accomplish our goals,” said Mattela. “The challenges this market faces with ‘reliable performance computing’ cannot be solved with existing architectures, but rather needs a completely new architecture built specifically to provide reliability, security, energy efficiency, and scalability. We strongly believe that building a technology foundation is key to developing highly differentiated products that can lead the industry.”
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Learn more about membership.