Robert Dennard, who invented fast memory technology still at the heart of computing and who captured the chip industry's march of progress in an idea called Dennard scaling, has received the prestigious Kyoto Prize.
In a ceremony Sunday, the Inamori Foundation awarded Dennard, an IBM fellow, the Kyoto Prize for advanced technology for the 1968 invention of DRAM. The award comes with $500,000.
Dynamic random access memory, or DRAM, is able to store information so a chip can retrieve it rapidly. It's central to the operation of everything from mobile phones to mainframes. Indeed, it's become more important in some areas: some companies house databases entirely in DRAM memory for a performance boost over conventional hard drives.
However, DRAM has its issues. For one thing, it draws a lot of power, and that's a problem when data center efficiency is crucial and mobile devices are limited by batteries. For another, its response speed hasn't kept up with speed improvements in microchips, requiring processor and computer engineers to insert various levels of faster memory between it and the DRAM to keep the chip supplied with data.
Dennard, now 81 and still working at IBM, also is known for what's called Dennard scaling, which encapsulated the wonders of chip improvements from the glory days of the semiconductor industry. Dennard noticed that as chips were made with electronics at ever smaller scales, the increases in the numbers of transistors was offset by the decreases in each transistor's power consumption. And because the smaller transistors generally would switch on and off faster, that meant chips would run at higher clock frequencies.
It wasn't technologically easy coming up with ever-smaller manufacturing process technologies, but it paid off with an explosion in computing performance. Alas for the computer industry, those relatively easy gains have come to an end, limited by power-consumption problems that have stalled the increases in chips' clock frequency.
"It went on for more than three decades. It was really great," said Sam Fuller, chief technology officer of Analog Devices, speaking of Dennard scaling in an earlier interview. "What you got with each generation was twice the transistors and an increase in speed and performance, with no increase in power consumed and no increase in cost."
With the end of Dennard scaling, software stopped getting faster automatically. It hasn't brought an end to Moore's Law, which says the number of transistors on a chip increases every two years but doesn't say anything about clock speeds. Nowadays, though, there's an emphasis on parallelizing software, which means breaking it down into independent elements that run in parallel on all those transistors.