Crossbar nanowire chips combine to form tiny CPU for beyond-Moore’s-law electronics

• By Ryan Whitwam on February 3, 2014 at 9:11 am
• • 
  • • As transistor technology continues its march forward with smaller, faster components, we’re getting ever closer to the point at which the realities of atomic scale will put an end to Moore’s law — unless we find a way around it. A team of researchers from Harvard and non-profit research company Mitre have devised a possible solution to the problem using nanowires as a stand-in for traditional transistors in tiny processors.
      The device created in the lab is by no means a match for modern computer processors, but it is built on a completely new process. The chip designed by chemist Charles Lieber and his team uses germanium core nanowires just 15 nanometers wide. The wires themselves are coated in silicon and are laid out in parallel on a silicon dioxide substrate. Embedded in the surface of the chip is a network of chromium and gold contacts, but these run the opposite way, creating a crisscross pattern.
      Each of the points in the chip where the nanowire crosses the embedded contacts can act as a programmable transistor node. Applying voltage to the nanowires toggles them between on and off. The researchers call this a “crossbar array.”
      The Harvard chip has 180 of these faux-transistors divided into three separate tiles. One tile is used to run basic mathematical operations and the other two store one bit of memory each. That makes this chip a simple 2-bit adder without any regular CMOS transistors. Yes, it’s a far cry from all but the most primitive CPUs, but the team believes this design can be scaled up simply by adding more nanowires to a larger grid of contacts. Four tiles would create a 4-bit adder array, for example.
      This isn’t the first time nanowires have been investigated as a way to circumvent the limits of Moore’s law, but the issues inherent with material at this scale have prevented it from being practical. Placing nanowires with the necessary level of precision is extremely difficult, and if a wire comes in contact with another one, it shorts out and knocks out all the transistor nodes down the line. Lieber and his team solved this problem with a technique dubbed “deterministic nanocombing.”
    • 
    • Before applying the nanowires, the substrate that will form the base of the chip is coated in a thin film of photoresist. Next up, narrow slots are carved out using electron-beamlithography. The slots are where nanowires are intended to go, but they won’t just slot themselves in. They almost do, though. The wires (which have already been grown on a different substrate) are chemically treated so they will stick to the exposed silicon oxide surface in the slots. Then the nanowire-encrusted substrate is dragged across the chip and the wires are deposited. The rest of the resist can be removed after the nanowires are situated.
      The circuits built with this process are small and very low-power, which makes them ideal for implantable devices like real-time biosensors. Imagine a tiny device that could be implanted under the skin to monitor blood glucose levels in diabetic patients, but uses virtually no power. The same properties could make nanowire chips perfect for advanced microcontrollers in robots. Lieber doesn’t see nanowires as a replacement for transistors in large-scale CPUs, but as a way to make processors far smaller and faster than silicon could ever scale.