Semiconductor Physics: Part 8 - Lithography
Welcome back, semiconductor fans! In the last installment, we introduced metal organic chemical vapour deposition (MOCVD) as a growth process to rival molecular beam epitaxy (MBE) in terms of industrial scalability, and we saw that some really quite scary gases are involved! We also saw an unexpected, novel application of MOCVD, whereby III-V nanowires of excellent quality can be grown (albeit randomly) by the introduction of gold (Au) nanoparticles to the surface. We noted that so far, we have been considering 'bottom up' growth processes, where nanostructures are built up, atom by atom, before discussing the characteristics of a cleanroom, in which semiconductor device fabrication takes place. This post will focus on the 'top down' approach to semiconductor device fabrication.
Lithography
At its core, lithography is really quite a simple process, that can be summarised broadly as the transfer of a pattern to a surface. We will begin our discussion by considering basic-contact photolithography. We start with our semiconductor substrate (or semiconductor epilayer), which could be a wafer that has just come out of an MBE machine. A very thin polymer layer is coated onto the wafer, in a spinner. A glass plate is held above the sample, on which there is some metal - normally chrome, which is formed into the pattern you want to transfer onto the wafer. Then you shine UV light on the sample, which has sufficient energy to break the crosslinked bonds of the polymer. This occurs only in the locations where the light goes through the metal - elsewhere, it is reflected. So the polymer layer on the surface of semiconductor has been changed chemically, in the locations where the light can get to it. In exactly the same way as an old fashioned photographic process, the regions where the crosslinks are broken can be 'developed away', leaving us with a pattern transferred by the mask to the surface of the semiconductor, with the pattern consolidated in a hard plastic, after the solvent is driven off.
The wafer is held down and spun at a tremendous rate (2000 - 5000 rpm), and a small liquid drop of the polymer is dispersed on the surface of the surface of the wafer. What we're aiming for at this stage of the process, is for maximal uniformity and homogeneity (we don't want variations in density). Humidity can be a problem, since if the material starts to dry, variations in composition and density can arise. We also require that there are no contaminants - it is very important that this process takes place in a clean environment.
So we now have a wafer with a polymer pattern on the surface. We can coat with a metal or dielectric, or the semiconductor can be etched. Everything that follows this requires that the pattern has been faithfully transfered from the mask to the surface of the polymer. Note that there is great potential for alignment issues - certain transistor processes require 35 aligned patterns. If the sample surface is not flat, lithography is really difficult. We rely on having planar, flat surfaces. Circuits are built up by multple process steps, so it's a challenge to align patterns from one process to another. There are two options: either try to make your design alignment tolerant or you need to have truly excellent alignment. There's quite a disparity between the research level where you are basically looking down a microscope and moving a mask manually, and the production level, where you have automatic pattern recognition systems that can align these things to within fractions of microns.
Now, I hope you recognise that like any other type of electromagnetic radiation, UV light has a wavelength. Once the size of the aperture the light is going through is around half the wavelength, it no longer possible to get any light through. The wavelength of incident light represents a resolution limit. For research purposes, the resolution limit is of the order of a micron. The UV light just comes from atomic transitions within a mercury discharge lamp. As you get to shorter and shorter wavelengths you can get the light through smaller and smaller apertures, but this gets expensive at an alarming rate!. Really expensive systems in the extreme UV (assuming the chemists have developed sufficient photo-sensitive polymers!) use KrF and AF excimer lasers, which can achieve wavelengths of 248 and 193 nm, respectively. The lithography tool used in intel's i7 chip fabrication process, for example, costs 1.4 billion USD! That's just the lithography tool!
We have so far described contact printing, where the mask is in direct contant with the sample, giving rise to a pattern transfer ratio of roughly 1:1 in terms of size. There are more expensive systems such as projection mask aligners (steppers) that facilitate better resolutions, since they boast a lensing system which acts to focus the light through the mask, onto the sample, with the potential to reduce the size of the pattern. These steppers are really expensive - Southampton University in the UK is one of the few that has one.
Thanks for reading folks, that's all for this post. Next time, we will do a bit of maths to demonstrate typical resolution limits, before moving on to consider electron beam lithography, which you can read a bit about here.
I thought I might mention that if anyone has any questions (on any of my posts), please feel free to ask in the comments and I'll do my best to get back to you!