Although the most famous law of computing is Moore’s, known to all and recently commented by Intel in a rather surprising script twist, there is another of which little is commented. Dennard scaling is a fundamental part of the history of architectures, chips, and transistors, where it is closely tied to Moore and at the same time widely unknown.
Dennard’s Law is known by various names that misrepresent its creator, Robert H. Dennard , as it is normally referred to as the MOSFET Escalation Act or simply the Dry Escalation Act. Obviously, it has the peculiarity that it was formulated for MOSFET as such, but over time it has been applied to any chip.
Dennard scaling, between density, price and performance
Although it may not seem so, Moore’s Law is directly related to Dennard and Pollacks , since between the three they make up and close the circle of current transistors and the industry. Each focuses on two global and general values that complement the circle, where, for example, Moore referred to the integration and price of wafers and transistors, Dennard refers to price and performance in his law.
Dennard established that as transistors reduce their physical size, the power density remains constant, or what is the same, the energy use is kept in proportion to the area of the chip because both the voltage and the current they also decrease with length.
And this is where it links to Moore’s Law, since if, as this one stated, each generation doubled the density of transistors, a circuit could become 40% faster and the energy consumption would remain the same.
The price and performance ratio is in check
Now we understand why Intel talks about price and performance when he mentions Dennard, since if a transistor could become 40% faster at each reduction while maintaining consumption, it means that the price / performance ratio would increase equally.
Although Dennard was right on this for years, we are currently reaching the nanometer limit and since 2007 there is speculation that this scaling does not work even in its theoretical form.
The scaling of complementary metal oxide semiconductors (CMOS) has not yet been done as such (at least according to Intel) and continuous progress can be seen as the ability to control manufacturing improves. They are not as limited by physics as by their ability to manufacture in large volumes with high precision. It is difficult, but this keeps costs, the problem is as we all know, leakage currents.
This years ago it had a more than expected turn towards the multi-core and later the multi-chiplet, something that Intel clings to to ensure that the performance per watt will grow to levels of the late 90s, where it should approximately double every 18 months. this ratio.