Blows Against the Empire: Lidow's Radical GaN Logic
Efficient Power Conversion (EPC) CEO Alex Lidow tells Light Reading he might be within two years of a breakthrough that would upend the entire semiconductor market: CMOS logic implemented in gallium nitride (GaN).
Lidow has been talking about the end of the silicon era for years. The claim has often been taken as hyperbole because, while GaN has much to recommend it, there was always a catch: the fact that nothing can compete with CMOS logic, and GaN simply can't do CMOS.
On the plus side, GaN has properties that make the performance of GaN power ICs (field-effect transistors, amplifiers, drivers, controllers, etc.) undeniably superior to their silicon counterparts in many circumstances. And silicon simply craps out in several situations, like, for example in high-frequency applications. You have to use GaN, or some other material.
For several reasons, economics favors GaN. This has led to a thriving market for GaN circuitry and components. EPC, Anadigics Inc. (Nasdaq: ANAD), Qorvo Inc. , Macom , Texas Instruments Inc. (NYSE: TXN), Intel Corp. (Nasdaq: INTC) and IBM Corp. (NYSE: IBM) are among the companies making GaN products or exploring the technology. (See Yes, We GaN! New Chips for Cable Networks.)
Silicon's hegemony in memory ICs is also being threatened. This summer, Intel and Micron announced a new memory, said to be 1,000 times faster than extant NAND memory, based on some unidentified "unique material compounds," which is to say, either not silicon or not solely silicon.
Meanwhile, the industry has been pushing silicon closer to its physical limits, to the point that speculation that Moore's Law might one day give out has leaked into even the popular press.
That might happen, but people working with silicon are confident it isn't happening soon. And even if the GaN power ICs and non-silicon memories take over in those markets, that would hardly mean the end of silicon. Again, that's because of CMOS logic, and GaN's inability to do CMOS.
Or so most people think.
Lidow isn't certain himself, but he was confident enough in EPC's prospects to tell Light Reading he'll know within two years whether or not he can implement CMOS logic circuitry in GaN.
Even while expressing uncertainty, he still couldn't help indulging in a bit of bluster, referring to CMOS logic as silicon's Alamo. He knows you know what happened at the Alamo.
If he succeeds, it could indeed mark the beginning of the end of the silicon era. He's cautious about providing too many details, but this is how he explained the challenge:
"Digital chips are CMOS -- complementary MOS, which means they have an electron-dominated transistor, and above it a hole-dominated transistor. They work as a pair; that's where 'complementary' comes from. You have a p-channel device and an n-channel device."
"In gallium nitride, electrons go like greased lightning, and holes don't go anywhere at all. Today, you can't make CMOS logic. You can make NMOS logic."
"But there are several approaches for gallium nitride that look very promising for making these positive charges -- holes -- transport very efficiently on the same chip. If we can overcome that challenge economically -- and I know pretty precisely what it will take -- if we can do it, then CMOS will also be superior in gallium nitride than in silicon, and the entire digital world will start transitioning to gallium nitride."
The ramifications for any application that requires pure speed (e.g., communications networks, data centers, supercomputing) would be profound. GaN is way, way faster than silicon -- theoretically as much as 1,000 times faster.
— Brian Santo, Senior Editor, Components, T&M, Light Reading