Researchers from IBM Corp. and the Georgia Institute of Technology in Atlanta have developed the first silicon-based chip capable of operating at frequencies above 500 GHz — 500 billion cycles per second — by cryogenically freezing the chip to 451 degrees below zero Fahrenheit (4.5 Kelvins).
By comparison, 500 GHz is more than 250 times faster than current PC CPUs, which typically operate at approximately 2 GHz. Computer simulations suggest that the silicon-germanium (SiGe) technology used in the chip could ultimately support even higher (near-TeraHertz — 1,000 GHz) operational frequencies even at room temperature.
The experiments, conducted jointly by IBM and Georgia Tech researchers, are part of a project to explore the ultimate speed limits of silicon-germanium (SiGe) devices, which operate faster at very cold temperatures. The chips used in the research are from a prototype, fourth-generation SiGe technology fabricated by IBM on a 200-millimeter wafer. At room temperature, they operated at approximately 350 GHz.
“This groundbreaking collaborative research by Georgia Tech and IBM redefines the performance limits of silicon-based semiconductors,” said Bernie Meyerson, vice president and chief technologist, IBM Systems and Technology Group, in a statement. “IBM is committed to working closely with our academic and industry partners to deliver the insight and innovation that will enable a new generation of high-performance, energy efficient microprocessors.”
SiGe is a process technology in which the electrical properties of silicon. The material underlying virtually all current microchips, and it is augmented with germanium to make it operate more efficiently. SiGe boosts performance and reduces power consumption in chips that go into cellular phones and other advanced communication devices. Until now, only integrated circuits fabricated from more costly III-V compound semiconductor materials have achieved such extreme levels of transistor performance, IBM said.
Dean Freeman, a lead analyst on silicon chips for Stamford, Conn.-based Gartner Dataquest, noted at room temperature the chip runs at 350 GHz. Given that fact, he said it’s conceivable within the next decade we could see devices running at the 500 GHz or even 1,000 GHz.
“It depends on a number of factors,” he said. “Though it operates at 500 GHz it requires liquid nitrogen for cooling. There could be a breakthrough over the next 10 years and we could see something else take place.”
Silicon-germanium technology has been of great interest to the electronics industry because it allows substantial transistor performance improvements to be achieved while using fabrication techniques compatible with standard high-volume silicon-based manufacturing processes. By introducing germanium into silicon wafers at the atomic scale, engineers can boost dramatically performance while retaining the many advantages of silicon.
Rob Enderle, principal analyst, The Enderle Group in San Jose, Calif., described the IBM announcement as a science experiment. It isn’t likely anyone is going to be able to use this capability, given the extreme cooling requirement, outside of labs for some time.
“What this does show is what will be possible in the next few decades and is like a proof of concept which tests the limits of the technology. In addition this is a showcase for IBM and Georgia Tech both of which are showcasing their capabilities in this competitive area of research and development,” he said. “This kind of performance will undoubtedly form the foundation, and once it is mature, for a new class of ever more intelligent computers and systems.”
At these speeds, Enderle told eChannelLine, and with the proper programming and software, there is hardware potential to create large, and scale highly accurate (virtually perfect) artificial reality and artificial intelligence that could, in many ways, exceed what humans are capable of.
“In fact, this performance probably exceeds by some magnitude what would be required for either with efficient programming suggesting that, by the time this is commercialized, we could at a single system being able to simultaneously emulate multiple environments and people relatively accurately,” he said. “This probably creates a window of somewhere between 15 and 30 years when things we have only imagined we could do will become possible.”
Officials said the research would be published in the July issue of the journal IEEE Electron Device Letters.