A new startup company based in Germany, Samronis developing what it calls “3D” chips to run AI models natively on smartphones, earbuds, VR headsets and other mobile devices.
The Cimron chips, co-created by Kai-Uwe Demasius and Aaron Kirschen, engineering graduates from Dresden University of Technology, use electric fields to perform calculations instead of electrical currents, which is the method of traditional processors. Kirshen claims that this enables chips to achieve higher energy efficiency while keeping the manufacturing costs of producing them low.
“Because of the expected shortage of AI computing resources, many companies with a business model that relies on access to such capabilities are risking their very existence — for example, large startups that train their own models,” Kirschen told TechCrunch in an email interview. . “The unique features of our technology will enable us to reach the price of today’s chips for consumer electronics even though our chips are capable of running advanced artificial intelligence, which others do not.”
Cimron chips—for which Demasius and Kirschen filed an initial patent in 2016, four years before Cimron was founded—exploit a somewhat unusual component, known as a “memory capacitor,” or a capacitor with memory, to power calculations. The majority of computer chips are made of transistors, which unlike capacitors cannot store energy; They act only as on/off switches, allowing or stopping electrical current.
Simron memory capacitors, made from conventional semiconductor materials, work by exploiting a principle known in chemistry as… Charge shielding. Memory capacitors control the electric field between the upper electrode and the lower electrode via a “protective layer”. The shielding layer is in turn controlled by the chip’s memory, which can store different “weights” of the AI model. (The weights essentially act like handles on the model, manipulating and adjusting its performance as it trains on and processes the data.)
The electric field approach reduces the movement of electrons across the chip, reducing power and heat usage. Semron aims to take advantage of the heat-reducing properties of an electric field to place up to hundreds of layers of memory capacitors on a single chip, dramatically increasing computing power.
“We use this property as an enabler to spread computing resources several hundredfold across the fixed silicon area,” Kirschen added. “Think of it like hundreds of chips in one package.”
In 2021 Stady Published in the magazine nature electronics, Researchers at Semron and the Max Planck Institute for Fine Structure Physics have successfully trained a computer vision model with a power efficiency of more than 3,500 TOPS/W – 35 to 300 times higher than current technologies. TOPS/W A little bit of Fuzzy metricBut the bottom line is that memory capacitors can lead to significant reductions in power consumption while training AI models.
Now, it’s early days for Semron, which Kirschen says is in the “pre-product” stage and has “minimal” revenue to generate. The hardest part of scaling up for chip startups is often mass manufacturing and reaching a meaningful customer base — though not necessarily in that order.
What makes things more difficult for Semron is the fact that it has stiff competition in custom chip projects 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 instead of transistors, but with a different substrate architecture.
However, Semron – which has a workforce of 11 people and plans to grow by around 25 people by the end of the year – has managed to attract funding from investors including Join Capital, SquareOne, OTB Ventures and Onsight Ventures. To date, the startup has raised €10 million (about $10.81 million).
SquareOne partner George Stockinger said via email:
“Computing resources will become the ‘oil’ of the 21st century. As large, infrastructure-hungry language models conquer the world and Moore’s Law reaches the frontiers of physics, a massive bottleneck in computing resources will shape the coming years. Inadequate access to computing infrastructure will It significantly slows down the productivity and competitiveness of both companies and entire nation-states. Semron will be key to solving this problem by providing a revolutionary new chip that is inherently specialized in computational AI models. It breaks with the traditional transistor-based computing model and reduces costs and power consumption for the task Specific computation by at least 20 times.