“ If the forecasts for the growth of global AI usage prove accurate and we continue using AI at the current rate, we would need to install roughly 400 additional nuclear power plants. There are around 400 nuclear power plants in operation worldwide today, so this gives you an idea of the scale involved. Alternatively, we’ d need to cover an area equivalent to the entire island of Mallorca with solar panels simply to generate the additional energy required,” Michael reveals.“ At the moment, our focus is on increasing the efficiency of converting high voltage into low voltage as required by the chip. We’ re moving towards higher voltages because it allows us to operate at lower current, reducing overall electrical losses.
“ At the same time, we are trying to operate at higher temperatures and higher frequencies, while continually improving efficiency. The shift toward silicon carbide and gallium nitride semiconductors, which can operate at higher temperatures and frequencies, also requires changes to the packaging materials surrounding them. If the semiconductor operates at a higher temperature, the other materials in the package must also be capable of withstanding those temperatures. As a result, we are moving from soldering to sintering, from DCB metal-ceramic substrates to AMB metal-ceramic substrates, and from aluminum wires to copper wires. In other words, the packaging materials themselves are what enable the efficiency gains and higher operating temperatures.”
Although improving the efficiency and performance of the underlying power electronics will be essential to limiting the exponential energy footprint of AI and data centers, semiconductor advancement is only part of the solution. As Michael explains, the materials and packaging surrounding them must also evolve to support higher voltages, temperatures, and frequencies while minimizing electrical losses.
142