Today AI’s large language models are largely dependent on Graphics Processing Units (GPUs) which are expensive, in limited supply and use an enormous amount of energy. Nvidia dominates that market but the race for the chip of the future is on.
Q.ANT, a German photonic computing company which specializes in next-level, energy-efficient processor technology for AI and high-performance computing, is aiming to be one of the contenders.
Q.ANT, which recently secured a €62 million series AI round of financing, has developed a photonic AI accelerator that it says enables drastic energy savings and performance gains: For complex AI workloads, Q.ANT’s photonic processors can reach 50x better performance than state-of-the-art GPUs and a 30x better energy efficiency, says CEO and Co-founder Michael Förtsch, a speaker on a panel about NextGen computing moderated by The Innovator’s Editor-in-Chief at the DLD Future Hub: Impact of AI in Munich Sept. 10.
Why does this matter? The next wave of computing will be shaped by advances in simulation, image analysis, classification, and other complex applications – areas where existing chip technologies reach their physical and economic limits: performance stagnates, energy demand skyrockets, and the costs explode, says Förtsch. Photonic processing is poised to become a critical pillar of next generation computing because it excels in exactly these domains, he says.
“We have reached the end of what digital chips can do,“ says Förtsch, who earned his doctorate at the Max Planck Institute for the Physics of Light in Erlangen, where he received the Otto Hahn Medal for his achievements in quantum information processing. “If we want to exploit the full potential of AI including pictures, audio and cognition, we need to do way more advanced math. The only alternative is a different type of processor. We are stepping into the ring to provide it.“
Photonic architecture uses analog computation with light instead of translating complex mathematical functions into digital bits and transistor flips. Light-based technology can execute complex mathematical operations much more efficiently than GPUs, accelerating AI training by 50 times, he says. Light also mitigates the heat issues that affect electrical circuits and, by using fewer parameters per workload, results in high energy efficiency.
Other startups, including Lightmatter, Lightelligence, Luminous, and giant tech players such as IBM are also working on photonic computing. Q.ANT, which counts Hermann Hauser, a serial entrepreneur and founder of Amadeus Capital Partners, and Hermann Eul, a former Intel executive and member of the board of Infineon Technologies, Europe’s largest semiconductor manufacturer, among its advisors, says it is the first photonic computing company to have publicly presented a prototype solution that easily plugs into existing data centers with no need for changes to current hardware or software stacks.
The Stuttgart-based company‘s hardware, called the Native Processing Server (NPS), is now operational at the Leibniz Supercomputing Centre (LRZ) in Germany, marking the first real-world deployment of analog photonic co-processors in high performance computing (HPC), according to Q.ANT. It is currently considering which data centers to work with.
“Our goal is to propel the performance of the compute sector so that applications that have been out of reach can become a reality; improve the energy efficiency of the chips and help Europe build a complete AI processor by using retrofitted CMOS lines – a model that can be copy and pasted anywhere else globally,” says Förtsch.
Q.ANT is addressing three pressing problems in the field of computing and AI:
- Traditional semiconductor technology – Complementary Metal-Oxide-Semiconductors (CMOS)- is becoming economically infeasible as the requirements of AI processing skyrocket.
- The energy consumption of AI processing threatens to destroy its business case.A graphics processing unit (GPU – the current semiconductor technology used for machine learning) requires 1,2 kW of energy per card: the energy consumption/requirement equivalent of a kitchen oven. A report from the International Energy Agency (IEA) estimates that by 2026, data center energy use will surpass the entire annual electricity consumption of Japan. “At some point somebody has to pay for all this energy,” says Förtsch. “If we continue like this, the business case is gone – and more importantly, we burn the planet. “
- A few companies in the USA and Asia control the AI industry, creating an imbalance of power. Q.ANT’s technology can be manufactured on repurposed 90nm chip foundries anywhere in the world. This democratizes the manufacturing process and access to this technology, he says.
The seven-year-old German scale-up operates its own chip pilot line in Stuttgart and manages the entire value chain in Europe – from wafer to chip design and industry-ready processor. Critically the company has demonstrated that this can be done by retrofitting existing chip factories, avoiding multi-billion-dollar investments in new facilities.
The scale-up’s pilot line in Stuttgart is producing 50,000 chips a year. The company is aiming for industrial scale in two to three years.”Classical CMOS is at the end of its life,” says Förtsch,“ so the time could not be better for Europe to innovate.“
This article is content that would normally only be available to subscribers. Become a subscriber to see what you have been missing.
