The two elements which have most supported economic development over the past fifty years are the standardization of maritime transport – thanks to the container which has opened up employment areas that were previously closed to the world – as well as the miniaturization of transistors and their concentration in a circuit integrated. We often ignore to what extent this miniaturization has irrigated all sectors of the economy. The increase in computing power for ever lower costs would represent half of the increase in productivity for oil exploration in the energy sector and 94% of the improvement in the speed of decoding proteins produced by our genes in the biotechnology sector at the source of new drugs and therapies.
This means that the impact of computing speeds is not limited to ever more powerful smartphones, but that much of the economic growth of the past fifty years was a second-order effect driven by Moore’s Law. However, this exponential growth will soon no longer be tenable. In any case, this is what several experts in the sector warn. In 1965 Gordon Moore predicted that the number of transistors in a dense integrated circuit would double every eighteen months, a figure he later revised to two years. In 1968 he co-founded Intel, and his observation became the company’s obsession, taking with it the rest of the semiconductor industry.
Validated by facts, it has achieved the status of law. To achieve their goals, the founders played on two main factors: the size of the wafer on which the transistors are placed and the fineness of the engraving of the group of components which will represent a chip. Moreover, the capacity for innovation of these actors is generally linked to the finesse of the engraving they can achieve.
The Death of Moore’s Law
3 nanometer chips are the new horizon of the sector. The South Korean Samsung drew the first last summer, quickly followed by the world leader, the Taiwanese TSMC, which devoted nearly 55 billion euros to it. It is difficult to grasp the scale of such small transistors: 1 nanometer is equivalent to 1 billionth of a meter; for comparison, a human hair is about 50,000 nanometers thick.
Giants love to project themselves into the future. Even though the 3 nanometer process has barely begun its industrialization phase for Apple, TSMC is already talking about 2 nanometer chips for 2026. These new miniature chips are essential to accelerate the transformation of the automotive industry towards the autonomous vehicle. To give an idea, the mega-factory in Dresden in Germany for which TSMC is in discussion should produce mature chips for car manufacturers of 22 and 28 nanometers, which are capable of supplying the tenors of the European sector Infineon, NXP and STMicroelectronics. The European Commission is pushing for this future site to also be able to go down to lower scales.
But this race for miniaturization is colliding with the laws of thermodynamics, which impose strict limits on the efficiency of certain processes. The Landauer limit corresponds to the tiny quantity of heat dissipated during each calculation operation. Add to this the current leakage, and the transistors, too close to each other, get too hot to work.
In December 2022, Nvidia CEO Jensen Huang even took the stage to declare Moore’s Law dead. To try to maintain its exponential growth trajectory, the chip industry will switch to three-dimensional architectures, as it already does for RAM, but this does not solve the problem of heat dissipation. More and more energy should be devoted to cooling these chips, even though energy will remain constrained due to the environmental transition. If quantum computing is sometimes presented as the rebound solution, its experimental nature calls for caution on the horizon of its development on an industrial scale. A wall seems to be looming in the 2030s. The performance boost won’t go away immediately, but it’s going to take some getting used to in a world where the main source of lower prices is drying up.
* Robin Rivaton is CEO of Stonal and a member of the scientific council of Fondapol