Nanowires are a perennial IEDM topic, and this year was no exception. Three papers in particular were identified as standing out.
One of them relates to efforts to work germanium into the mix for pFETs. Such work is all about mobility, and a team from MIT achieved twice the mobility of biaxially-strained planar Si. This was done using biaxially-strained germanium that was then patterned into wires using e-beam lithography, which relaxed the lateral strain. The result wasn’t uniaxial strain, but asymmetric. HfO2 was used as the high-κ dielectric, which also acted as an etch-stop when the nanowires are being formed as well as passivation for the Ge/dielectric interface.
A team at Purdue, meanwhile, investigated III-V nanowires for nFETs using InGaAs. They used 20-nm nanowires that were actually a sandwich of In0.53Ga0.47As between In0.65Ga0.35As for higher mobility and lower interface defect density.
They ran three different gate stacks. Two of them used 0.5 nm Al2O3 and 4 nm LaAlO3, with one reversing the order of the stack as compared to the other; this was surrounded by WN. The EOT* of these was 1.2 nm. The other “stack” was simply 3.5 nm of Al2O3 (andthe WN); it had an EOT of 1.7 nm.
The resulting structures exhibited a subthreshold slope (SS) of 63 mV/dec and DIBL of 7 mV/V; Ion was a strong 0.63 mA/µm and gm was 1.74 mS/µm. The SS and gm are the best yet reported; short-channel effects were negligible.
Finally, a team at the Swiss Federal Institute of Technology in Lausanne (EPFL) experimented with ambipolar nanowire structures – devices that can be switched in real time to behave as n-type or p-type. While seen as an annoyance conventionally, this project leveraged the phenomenon by creating a “stack” of four nanowires vertically (using DRIE) and then forming two gates. In the center of the wire was the “standard” control gate; contacting both ends near the source and drain was the polarity gate.
The critical thing about this was that the voltages used to control the two gates were roughly the same. This creates the potential for using both gates in logic designs, the natural function being the XOR gate (with echoes of using MRAM cells as XOR gates). It’s suggested that using XOR gates instead of inverters/NAND/NOR gates can reduce the resources required, although obviously the basic logic math changes due to the different primitive function.
If you have the proceedings, the MIT paper is #16.5; the Purdue paper is #27.6, and the EPFL paper is #8.4. (Yeah, I know, all different sessions… how things get grouped at IEDM remains a mystery to me, but it seems to work for them…)
*EOT is “equivalent oxide thickness.” The whole idea of high-κ materials is to provide the “reactivity,” if you will, of a super-thin layer of SiO2 without all the electrons tunneling through because it’s so thin. So you get a thicker layer of material that acts like a layer of SiO2 having a thinner EOT. So this allows the thicknesses of different materials to be “normalized” to SiO2-equivalent thicknesses.