Chip innovations are getting smaller and more powerful
Computer chips are the foundation of our digital world. No smartphone, no computer, no car and no remote control do without them. And one not so distant day will probably be installed in our head.
Above all, artificial intelligence (AI) demands the demand for more computing power in a confined space. The chip manufacturers reach physical limits. European start-ups and universities are researching chip innovations to breastfeed the growing hunger for computing power.
In 1971 the chip company "Intel" with its "4004" launched the first commercially successful microchip. At that time it was a sensation, on the other hand, the three-time four millimeter arithmetic chip looks chunky. And above all underpowered: they brought just 2,300 transistors - unimaginably a lot for then, ridiculously little for today.
To this day, transistors form the heart of every computer chip. Tiny switches that switch between zero and one back and forth between on and out. These switches lay the foundation of our digital world. The more transistors - the more powerful the chip.: And the global demand for computing power and data storage explodes.
The chip industry has been working above all for years: How many of these transistors still fit on a chip? Over the decades, transistors have become smaller and smaller, so that they only consist of a handful of atoms today. Today these switches are thinner than human hair, smaller than a red blood cell and provided with kilometers of wiring. A tiny switch-500,000 times smaller than a millimeter-which has now become absolutely irreplaceable for our daily life. 200 million transistors can find space on a square millimeter, and even several ten billion on a chip
But in the near future, this shrink attempt by the semiconductor science reaches its physical border.
We are currently using chip technologies such as "CPUS-Central Processing Units" for computers and smartphones. The alternative technology of the GPUS Grafics Processing Units-also called graphics cards, were originally developed for images, video content and 3D graphics for computer screens. Chip manufacturers »Nvidia« made a name for itself with these chips and is currently riding the wave of demand for AI chips. Because these graphics cards have the advantage that they can carry out parallel tasks and perform many tasks at the same time - and that is exactly what a AI needs so that they can work efficiently.
For AI, graphic chips proved to be the best solution currently, GPUs are the emergency solution for AI algorithms. Because at the moment there are simply no new approaches for AI chips. What there is is a lot of research and innovations. Although every nanometer is packed with switches on your chip, there is still unused space, in height. "Semron" researches this. The Dresden start-up developed chips that bring AI directly to end devices such as smartphones and headphones. In this way, data can be processed locally on the devices, which offers an advantage, especially with sensitive information. So that the chips become efficient enough, the chips must be as small, compact, inexpensive and energy -efficient
Co -founder Aron cherries are clear that it is not enough to put three, four or five chips on top of each other in a housing. The five-time performance combined with the costs for five chips does not help if a 1000x of current performance is necessary. Instead, Semron plans to raise several chiblages on top of each other during the manufacturing process. The patented semiconductor technology "Capram" allows you to process AI models locally. With memory chips as we know them from smartphones, it works. Up to 200 locations of savings are located on chips in our cell phones. In processors - i.e. the computers of a chip - this technology proves to be more difficult. Firstly, because the transistors are not so easy to stack on top of each other. And on the other hand, because this construction needs more energy and threatens to burn the chip with higher energy density. "Semron" boasts of solving this problem. The team now faces the challenge that the chip manufacturers also implement the patented manufacturing process and produce the chips in a large batch. The idea alone is not enough to satisfy the ever greater hunger for computing power. Because the chips also have to be produced - and this does not happen in Germany and not in Europe. It is comparable to the industrial revolution 150 years ago.
