ARM has published a new article on the Arm Blueprint blog, claiming that the computing paradigm has changed. The company emphasizes that semiconductor designs should replace Moore’s Law with the principle of “performance per watt”.
We often talk about transistor density in chips. According to Moore’s Law, “the number of transistors in microprocessors must double every two years” for continued performance improvement in our devices. However, if we look at the developments in the last few years, this will not be easy.
As semiconductors continue to shrink, things will get incredibly difficult as the atomic size approaches the dollar-yellow and it will become impossible to build smaller circuits. Different techniques and technologies may be developed in the coming period, but there is a different point that ARM draws attention to.
According to ARM’s article, computational efficiency is one of the primary factors for the advancement of modern designs. That’s why performance per watt qualifies as a new paradigm to target. Engineers should aim to achieve a certain performance while using as little power as possible to achieve better designs.
According to author Rob Aitken, machine learning can be used for ultra-low power consumption and optimization of workloads. In addition, techniques such as TinyML, which consumes milliwatts of power, are gaining momentum.
Moore’s Law has long served as a roadmap for the tech industry. Aitken, on the other hand, says that this is not a natural law and is merely an observation of the direction of technology. To summarize, according to ARM, the semiconductor industry should focus directly on “efficiency” and focus on maximum performance at low efficiency.