Air-gap insulation method increases chip speed
Faster and more efficient microprocessors have been created, say researchers at IBM. The T J Watson Research Center in Yorktown, USA, has unveiled a technique that uses self-assembly methods to create air gap insulation for wires in microprocessors. These insulators can increase the speed of a chip by 35% and consume 15% less power.
Smaller chips require copper wiring (of about 70nm width) that are fabricated close together. But the closer the wires are, the more likely their electrical currents are to interfere with each other. Insulation helps prevent this, but it can sap energy and slow data flow.
Traditionally, glass has been used as an insulator. But, Daniel Edelstein, IBM fellow and chief scientist of the project, explains that the relative dielectric constant (k) of glass of 4.0 is too high for microprocessors, which are becoming smaller and smaller.
‘Lower the k, and you lower the parasitic loss of electrical energy,’ he says.
Since air has the lowest possible dielectric constant of 1.0, it is the ideal choice for insulating copper wires.
To create air gaps small enough for microprocessors (35nm in diameter), IBM researchers have developed a self-assemblying material. An organic diblock copolymer (the exact composition of which cannot be revealed yet) is poured over copper wires that are embedded in a silicon wafer. The device is then baked at around 100-120ºC.
When heated, the polymeric molecules pull away from each other to form a regular array of uniform nanoscale holes, 20nm in diameter. These are used as a template to etch hollow columns into the glass dielectric surrounding the wires. Plasma is then pumped through the holes to blast away the remaining insulating material.
When the next level of glass insulator is deposited on the wires in a vacuum chamber, the air is removed, creating a vacuum between the wires, referred to as air gaps.
‘Since our gaps do leave some carbon-doped glass insulator above and below the interconnects, for mechanical strength and thermal conductivity, our effective dielectric constant is 2.0 – still a big jump [from the 4.0k of glass] in one step,’ says Edelstein.
‘Once the recipe for volume synthesis of this material was arrived at, we then treated this as a typical photoresist (except with no “photo”),’ he explains.
‘No problems have been found with shelf-life limits, trace metal contamination, particles, spin-speed process window, uniformity on 300mm wafers, or any other typical photoresist concern.’
Edelstein believes this technique will be of use to many different chips that require higher performance or reduced power consumption. ‘Lowering wiring capacitance is virtually always beneficial,’ he says.
|NPL image.jpg||140.55 KB|