A new Germany-based startup, Semron, is developing what it describes as “3D-scale” chips to run AI models locally on smartphones, headphones, VR headsets and other mobile devices.
Co-created by Dresden University of Technology engineering graduates Kai-Uwe Demasius and Aron Kirschen, Semron’s chips use electric fields to perform calculations instead of electric currents – the backbone of conventional processors. This allows chips to achieve higher energy efficiency while lowering the manufacturing costs to produce them, Kirschen says.
“Due to an expected shortage of computational resources for AI, many companies whose business model relies on access to such capabilities are risking their existence – for example, large startups that train their own models,” Kirschen told TechCrunch in an email interview. “The unique characteristics of our technology will allow us to match the price of current chips for consumer electronics devices, even though our chips are capable of running advanced AI, which others are not.”
Semron’s chips — for which Demasius and Kirschen first filed a patent in 2016, four years before founding Semron — leverage a somewhat unusual component known as a “memcapacitor,” or capacitor with memory, to perform calculations. The majority of computer chips are made of transistors which, unlike capacitors, cannot store energy; they simply act as “on/off” switches, either passing an electric current or stopping it.
Semron’s memcapacitors, made from conventional semiconductor materials, work by exploiting a principle known in chemistry as charge shielding. Memcapacitors control an electric field between a top electrode and a bottom electrode via a “protective layer”. The protection layer, in turn, is controlled by the chip’s memory, which can store the different “weights” of an AI model. (Weights essentially act like knobs in a model, manipulating and adjusting its performance as it trains and processes data.)
The electric field approach minimizes the movement of electrons at the chip level, thereby reducing power and heat consumption. Semron aims to exploit the heat-reducing properties of the electric field to place up to hundreds of layers of mems capacitors on a single chip, significantly increasing computing capacity.
A diagram showing the design of Semron’s 3D AI chip.
“We use this property to deploy several hundred times more computing resources on a fixed area of silicon,” Kirschen added. “Think of it as hundreds of chips in one package.”
In a 2021 study published in the journal Natural electronics, Researchers from Semron and the Max Planck Institute for Microstructural Physics have successfully trained a computer vision model at power efficiencies above 3,500 TOPS/W, 35 to 300 times higher than existing techniques. TOPS/W is a bit of a vague metric, but the takeaway is that memory capacitors can lead to dramatic reductions in power consumption when training AI models.
Today, Semron is in its early stages, which Kirschen says is in the “pre-product” stage and has “negligible” revenue to show for it. Often the hardest part of scaling a chip startup is mass manufacturing and reaching a meaningful customer base – but not necessarily in that order.
What makes things more difficult for Semron is the fact that it faces stiff competition from custom chip companies like Kneron, EnCharge and Tenstorrent, which have collectively raised tens of millions of dollars in venture capital. EnCharge, like Semron, designs computer chips that use capacitors rather than transistors, but using a different substrate architecture.
Semron, however, which has a headcount of 11 and plans to increase by around 25 by the end of the year, has managed to attract funding from investors such as Join Capital, SquareOne, OTB Ventures and Onsight Ventures. To date, the startup has raised 10 million euros (~$10.81 million).
SquareOne partner Georg Stockinger said via email:
“IT resources will become the “oil” of the 21st century. As large, infrastructure-intensive language models take over the world and Moore’s Law reaches the limits of physics, a massive bottleneck in computing resources will shape the years to come. Insufficient access to IT infrastructure will significantly slow down the productivity and competitiveness of businesses and entire nation states. Semron will be a key part of solving this problem by providing a revolutionary new chip that is inherently specialized in computing AI models. It breaks away from the traditional transistor-based computing paradigm and reduces costs and power consumption for a given computing task by at least 20 times.
Tech
