MIT World: Engineering >>
The world is counting on the fulfillment of (Intel co-founder) Gordon Moore’s Law for at least another half century. In Craig Barrett’s view, solutions to the crucial challenges of our time depend on improving on already nano-sized microprocessors every few years.
He points to the astonishing improvements in efficiency and miniaturization in Intel’s semiconductors, which around 1972 came loaded with 2,000 transistors that could be seen with the naked eye. Today’s integrated circuits, 11 generations down the road, bear 1-2 billion transistors that can be seen only with a scanning electron microscope. Intel has had to make other improvements too, says Barrett, as they moved into the nanoscale, attempting to improve functionality and performance without power dissipation. Dual and quad core microprocessors now permit parallel computing within a single PC. Barrett recounts how the first teraflop computer he worked on at Sandia Labs required 10 thousand Pentium processors and took up 2,000 square feet. “The challenge is in the next six to eight years, going to exascale, getting up to a million teraflops,” through multiple core processors, he says, and then there will be a “huge
He points to the astonishing improvements in efficiency and miniaturization in Intel’s semiconductors, which around 1972 came loaded with 2,000 transistors that could be seen with the naked eye. Today’s integrated circuits, 11 generations down the road, bear 1-2 billion transistors that can be seen only with a scanning electron microscope. Intel has had to make other improvements too, says Barrett, as they moved into the nanoscale, attempting to improve functionality and performance without power dissipation. Dual and quad core microprocessors now permit parallel computing within a single PC. Barrett recounts how the first teraflop computer he worked on at Sandia Labs required 10 thousand Pentium processors and took up 2,000 square feet. “The challenge is in the next six to eight years, going to exascale, getting up to a million teraflops,” through multiple core processors, he says, and then there will be a “huge