Intel co-founder Gordon Moore argued that the number of transistors on an integrated circuit would double every couple of years. And, to his credit, this rule was fulfilled during the time from 1965 to 2015, until the laws of physics began to interfere. Now researchers in North Carolina believe that we should not focus on further miniaturization of transistors in an effort to make chips even more efficient. Instead, they turned to chaos theory in the hope that the mixed structure would help provide performance gains that Intel cannot achieve.
Lead researcher Benam Kia explains that "the physical limit in the size of the transistor has been reached." If you listen to Intel presentations, you will notice that each new manufacturing process becomes more difficult to implement. It is not so easy to start mass production of perfect 14nm chips, and therefore the company had to postpone the production of 10nm chips several times. But Kia and colleagues believe the obsession with size has overshadowed a key fact about the design of modern microcircuits.
In a typical chip, there are a number of transistor circuits, each dedicated to a specific function. You can imagine a factory where each circuit is an employee with a calculator in his hands, who constantly solves the same equation every day. The first chips had a few employees, but as time went on, walls were demolished, calculators got smaller, and employees lost weight. This means that more employees can now huddle in the same building, but each of them can still perform the simplest math task when required.
In other words, a large number of transistors are dormant, creating a huge unused potential in the system. Kia explains that the new chip uses "chaos theory - the inherent nonlinearity of the system - to use programmable transistor circuits to perform various tasks." In the worker metaphor, the factory will abandon new employees and instead train existing ones to do multiple operations. This way more work / computation can be done with the same number of transistors / employees and apparently it is not that difficult to implement.
The researchers believe that although their idea is still purely theoretical, creating programmable transistor circuits will not be too difficult. In their opinion, these reconfigurable chips can be made using the same tools that Intel currently uses on its production lines. If this is the case, then we can expect an increase in processor performance, but for now we will have to wait for material scientists to learn how to make workable chips less than 5 nanometers in size.