Cornami, the startup with the many-core accelerator compute fabric, has received strategic investment into its over-subscribed Series C round from Applied Ventures, the venture capital arm of Applied Materials.
Cornami’s focus is hardware acceleration for fully homomorphic encryption (FHE) — an encryption scheme which allows data to be used while encrypted, without ever revealing the plaintext. The technology is still in its infancy today as FHE encryption algorithms are too compute intensive for practical implementation, but if practical FHE could be realized, there are huge implications for military, medical and other data which requires the highest level of secrecy. An FHE accelerator could enable using medical data from multiple sources to train a big AI model for medical imaging, for example, or transmitting and using military data without making it vulnerable by decrypting it at any point.
Describing Applied Materials as both an investor and a future customer, Cornami CEO Wally Rhines told EE Times that equipment manufacturers like Applied Materials would like to access data on their equipment as it is being used in customer facilities, but that this is not possible today.
“[Applied Materials’] customers don’t want their equipment manufacturers monitoring their equipment,” said Rhines. “They don’t want to send their sensor data to other people. They worry that if they give you enough statistical data to do meaningful things, that data will end up in the hands of their competitors.”
Using data on manufacturing equipment for analysis and diagnostics without ever decrypting it could mean equipment manufacturers can gain better insight into how machines are performing, whether they require maintenance, and perhaps, where any points of failure might be for particular models to avoid in future models under development. This type of data becomes much more valuable when it can be combined with data from many different customers.
“It’s great to have an investor who believes in the technology, who sees the business opportunity and the strategic advantage it provides, but is also in the business of investing in businesses they believe have a future impact on their [own] business,” Rhines said. “Applied Materials is an ideal partner.”
Funding and competition
Cornami’s Series C, led by SoftBank Vision Fund 2, was originally announced in October 2021 as a $50 million round; Rhines said the total today “far exceeds that amount”. Cornami was previously planning to have FHE accelerator silicon by 2021, but delays raising this funding meant a slowdown in development, compounded by the pandemic. The company today is back to around 54 employees (around the same number as when Rhines joined in March 2020), and is working towards its first tapeout.
While the amount invested by Applied Materials is undisclosed, Rhines said work on the Cornami chip is full steam ahead.
“We are funded to product success, with enough funding that we’ll either go cash flow positive and won’t need another round, or this will carry us into lots of customer ramp ups and evaluations,” Rhines said. “Either way, it gets us through the product development phase.”
While the funding delay has cost the company almost a year, Rhines is confident Cornami still has a sizeable head start on competing offerings.
“We have many years of verification on our TruStream computational fabric, which includes a compiler technology which allows you to divide an application up into totally independent streams of control and data, which have no interdependencies,” he said. “Performance is directly proportional to the number of streams, and adding more cores can scale performance linearly.”
While the initial FHE accelerator chip will have thousands of cores, future scaled down versions will go into markets like AI, autonomous vehicles and 5G communications.
Cornami faces competition from other companies working on FHE acceleration, including Intel (who is building a chip under the DARPA Data Protection in Virtual Environments program). According to Rhines, FHE is not suitable for GPU acceleration, so Cornami does not consider Nvidia a competitor in this space. Most of the work being done today on speeding up FHE is on the algorithm side, in companies Rhines counts not as competitors, but as future customers.
Cornami has also entered into a partnership with cybersecurity company Inpher, which powers privacy-preserving AI and analytics, primarily using multi-party computation (MPC) techniques today.
“Cornami’s unique and scalable computational fabric is the first compute platform we’ve seen that can meet FHE’s substantial computing requirements hitting both performance and cost requirements for a commercially viable product,” said Dimitar Jetchev, CTO and co-founder of Inpher, in a statement. “We look forward to working with Cornami to provide unrestricted FHE solutions to the market.”
Rhines said Inpher will use MPC in the short term, and both MPC and FHE alongside each other in the long term, eventually helping customers take advantage of Cornami hardware offerings.