Transistors measuring just 0.7 nanometer have been fabricated for the first time in electronics history. This size, equivalent to that of a glucose molecule, allows nearly 100 billion transistors to be installed on an area as large as a fingernail. A technological leap that pushes the boundaries of miniaturization.
Until now, going below the two-nanometer mark posed major problems. Electrons would become trapped by defects or escape from gates, leading to a loss of efficiency. These obstacles seemed insurmountable to continue making chips more powerful.
The solution devised by IBM is based on an architecture called NanoStack. Instead of arranging transistors flat, it stacks them vertically, like floors in a skyscraper. This technique, known as "nanostacking," uses an innovative dielectric bonding to connect each layer, allowing each transistor to be powered independently.
Initial tests show spectacular gains: a 50 % performance increase and a 70 % reduction in energy consumption compared to the best current 2 nm chips. SRAM memory, essential for fast calculations, also sees its density improved by 40 %, a major asset for artificial intelligence.
The future looks even more promising. Researchers estimate this technology can be pushed down to transistors of 0.1 nanometer, more than six times smaller. This would give a new lease on life to Moore's law, which predicts a regular doubling of transistor density.
According to IBM representatives, this architecture is not a simple improvement but a generational change. It should replace current technologies within five to ten years, becoming the new standard for processors and graphics chips.
Moore's law
This observation by Gordon Moore in 1965 predicted that the number of transistors on a chip would double every two years, leading to increased performance and lower costs. For decades, the industry followed this law, but with extreme miniaturization, physical limitations have emerged.
Quantum effects and current leakage become problematic below a few nanometers. Manufacturers have had to innovate with 3D architectures like FinFET, then nanosheets. IBM's NanoStack pushes the limits further, potentially allowing continued progress.
However, some experts believe Moore's law is coming to an end. Even if solutions like vertical stacking provide a respite, obstacles such as heat dissipation and process variation could slow progress. The NanoStack is thus a breath of fresh air, but not an eternal solution.
For now, this innovation keeps the pace, but the industry is already exploring other paths such as silicon photonics or neuromorphic computing to continue advancing beyond the limits of silicon.