Car, refrigerator, mobile phone, or credit card. All of these seemingly very different objects have one thing in common – they could not function without chips. Hundreds of billions of chips are produced worldwide each year, and each product requires a different type. Designing chips for specific applications is therefore quite complex. However, Czech experts have developed an algorithm that can complete a task that would normally take several days in just a few tens of minutes. It is already being used by a global multinational company.
“There are tens or even hundreds of chips in a single laptop. Sometimes, however, even one chip can do the work of dozens of chips,” explains Zdeněk Hanzálek, head of the Optimization Department at the Czech Institute of Informatics, Robotics and Cybernetics, CTU. A chip is a printed circuit of minuscule dimensions, often invisible to the naked eye, connecting components like transistors and resistors.
“Underneath the chip, there are, simply put, ‘legs,’ because you have to connect it to many interfaces. So the real complexity is hidden in those tens of nanometers and in the technologies used to assemble it,” adds Zdeněk Hanzálek, illustrating the complexity of this “puzzle” using an analogy of a city with houses and streets.
“If you look at a city from an airplane, you see streets, houses, stadiums, and some factories. If you zoom in on a building, you can distinguish individual machines and some communication lines showing how materials are transported,” he describes, explaining how one can imagine a chip.
From hours to minutes
The algorithm developed by CTU experts significantly simplifies chip design. It is enough to input requirements, taking certain constraints into account.
“First, we have to design the circuit, what it will contain,” demonstrates Josef Grus directly in a specialized program. “We can place some transistors into our design environment. For example, a transistor measuring 2 x 4 micrometers – there will be several copies, say eight, forming some logical structure.”
On the screen, these elements appear as red and green dots connected by blue lines.
“We draw the blue connections, indicating which transistor is connected to which, then we run the algorithm and wait for the components to be arranged,” continues the developer, starting the automatic process, which can take from a fraction of a second to tens of minutes, depending on the complexity of the unit being designed.
“Even for very small circuits, manual work takes tens of minutes or even hours, while the algorithm can find a good solution for additional circuits in minutes or tens of minutes, saving a really large amount of time,” emphasizes Josef Grus as the main advantage of the new algorithm.
In addition to optimizing chip design, it can also be used, for example, for arranging machines in a factory.
