Semron, a newly established startup based in Germany, is developing innovative integrated circuits described as “3D scaled” chips that enable running artificial intelligence models locally on smartphones, earphones, virtual reality glasses, and other mobile devices.
Kai-Uwe Demasius and Aron Kirschen, engineering graduates from Stuttgart, are behind the creation of Semron. The integrated circuits developed by the company use electric fields for computations instead of using electric current, which is typical for traditional processors. This allows Semron’s chips to achieve higher energy efficiency while reducing production costs, according to Kirschen.
“In light of the expected shortage of AI computing resources, many companies whose business models rely on access to such capabilities are risking their existence – for example, large startups that train their own models,” Kirschen said in an email interview with TechCrunch. “The unique features of our technology will enable us to achieve the price point of today’s circuits for consumer electronics, even though our chips can run advanced artificial intelligence, which others cannot do.”
Semron’s chips, for which Demasius and Kirschen filed the first patent in 2016, four years prior to founding Semron, utilize a slightly unconventional component known as a “memcapacitor,” a capacitor with memory, for conducting computations. Most computer chips consist of transistors, which unlike capacitors, do not store energy; they act as “on/off” switches, either allowing or interrupting the flow of current.
Semron’s memcapacitors, made of conventional semiconducting materials, operate by utilizing a phenomenon known in chemistry as “charge screening.” The memcapacitors control the electric field between the top and bottom electrodes using a “protective layer.” The protective layer, in turn, is controlled by the memory of the chip, which stores different “weights” of the AI model. (Weights essentially act as knobs in the model, manipulating and adjusting its performance during training and data processing.)
With their electric field-based approach, Semron’s chips minimize electron movement at the integrated circuit level, reducing energy consumption and heat. Semron plans to leverage the heat-reducing properties of the electric field to house hundreds of layers of memcapacitors on a single chip, significantly increasing computational power.
“We make use of this property as a tool to enable the placement of several hundred times greater computational resources on a fixed silicon surface,” Kirschen added. “You can imagine it as hundreds of chips on one chip.”
In a study published in 2021 in the journal “Nature Electronics,” researchers from Semron and the Max Planck Institute for Microstructure Physics successfully conducted training for a computer vision model using memcapacitors with an energy efficiency exceeding 3500 TOPS/W, representing a 35 to 300-fold increase compared to existing techniques. TOPS/W is a fairly general metric, but the essence is that memcapacitors can lead to significant energy consumption reduction during AI model training.
It is still early days for Semron, as Kirschen claims. The company is in the “pre-production” phase and has “negligible” revenue. Often, the most challenging part for a startup dealing with integrated circuits is mass production and acquiring a significant customer base, although not necessarily in that order.
Helpful aspects for Semron include the fact that it faces competition from similar integrated circuit projects such as Kneron, EnCharge, and Tenstorrent, which have collectively raised tens of millions of dollars in venture funding. EnCharge, like Semron, designs computer chips using capacitors instead of transistors, but on a different substrate architecture.
Semron, currently employing 11 people and planning to increase that number by about 25 by the end of the year, has attracted investments from entities such as Join Capital, SquareOne, OTB Ventures, and Onsight Ventures. So far, the startup has raised 10 million euros (~10.81 million dollars). Georg Stockinger, partner at SquareOne, commented on this in an email:
“Computational resources will become the ‘oil’ of the 21st century. With the growing need for massive language models and the physical limits reached by Moore’s Law, significant computational resource constraints will shape the coming years. Insufficient access to computational infrastructure will undoubtedly slow down productivity and competitiveness for both companies and countries. Semron will be a key element in solving this problem, delivering a revolutionary new chip specialized in AI model computations. It will break away from the traditional transistor-based paradigm and reduce costs and energy consumption for a given computing task by at least 20 times.”
FAQ:
1. What does the startup Semron do?
Semron is developing innovative integrated circuits enabling the running of artificial intelligence models locally on smartphones, earphones, virtual reality glasses, and other mobile devices.
2. What are the unique features of Semron’s technology?
Semron’s integrated circuits use electric fields for computations instead of traditional electric currents, increasing energy efficiency and lowering production costs.
3. How do Semron’s memcapacitors work?
Semron’s memcapacitors utilize the phenomenon known as “charge screening” for computations. They control the electric field between the top and bottom electrodes using a “protective layer,” which is controlled by the chip’s memory that stores different “weights” of the AI model.
4. What are the benefits of using Semron’s chips?
By utilizing electric fields, Semron’s chips minimize energy consumption and heat. Additionally, Semron’s chips have the potential to increase computational power by placing hundreds of layers of memcapacitors on a single chip.
5. What research has been conducted regarding Semron’s technology?
Researchers from Semron and the Max Planck Institute for Microstructure Physics successfully trained a computer vision model using memcapacitors with an energy efficiency exceeding 3500 TOPS/W, representing a 35 to 300-fold increase compared to existing techniques.
6. What are the prospects for Semron’s development?
Semron is currently in the pre-production phase and has negligible revenue. However, the company has attracted investments from various entities and plans to increase its staff by the end of the year.
7. How does Semron differ from competing projects?
Semron competes with other integrated circuit projects such as Kneron, EnCharge, and Tenstorrent. Unlike transistors, both Semron and EnCharge design computer chips utilizing capacitors.
8. How can Semron impact the future of computing?
Semron has the potential to deliver a revolutionary new chip specialized in AI model computations, reducing costs and energy consumption for a given computing task by at least 20 times.
The source of the article is from the blog trebujena.net