In my post a couple of weeks back on A Matter of Scale: Electromigration, I mentioned that once silicon features shrunk below 100nm into the world of nanotechnology, previously negligible physical phenomena now play a big role. One of those annoying phenomena is the well proximity effect (WPE).
To explain how WPE affects integrated circuit designers, we have to first talk just a little bit on how PMOS transistors are created. The figure above shows the standard textbook cross-section of a silicon wafer with an NMOS and PMOS transistor side-by-side. The silicon wafer, also called the substrate, is a crystalline structure where each silicon atom shares four covalent bonds with neighbouring silicon atoms. However, it is standard in the IC industry to lightly dope the silicon wafer substrate with boron. If you remember your high school chemistry, silicon has four valence electrons and boron has only three. When boron is introduced into the silicon crystal structure, the four covalent bonds it shares with neighbouring silicon atoms is short an electron. Hence, this type of material is called p-type silicon, where p stands for ‘positive’ (absence of electrons). The inverse to doping silicon with boron is doping with phosphorus, which has five valence electrons. When phosphorus bonds with four neighbouring silicon atoms, an extra electron is left over, hence creating n-type silicon; n stands for ‘negative’ (excess of electrons).
NMOS transistors can be built right into the p-type silicon substrate. PMOS transistors, on the other hand, need to be built into n-type silicon. By bombarding a small area at the top of the silicon wafer with phosphorus, a tub of n-type silicon can be created inside of the p-substrate. As the diagram shows, this is typically called the n-tub (or n-well). With the creation of the n-well, PMOS transistors can now be made.
During the fabrication process of ICs, one of the first steps is to create n-wells all throughout the wafer where PMOS transistors will live. First, silicon-oxide (SiO2) is grown on top of the silicon wafer.
However, the doping concentration is not uniform throughout the n-well. While the phosphorus bombardment is ideally perpendicular to the plane of the silicon surface, in reality, there is some angle to some of the particles. This means that phosphorus dopants get reflected from the wall of the SiO2 barrier, creating a higher concentration of dopants at the edges.
When transistors were larger in the 130nm node, the minimum allowable distance from the PMOS to the edge of the well was large enough so that transistors were always kept a safe distance away from the higher concentration of the well edge. However, in the 90nm node, dimensions got so small that the allowable distance between the well edge and the PMOS transistor is now encroaching into the area of higher dopant concentrations. This means two identically built PMOS transistors side-by-side will have different characteristics due to their differing distances from the n-well edge. This creates problems for circuits that rely on good matching between close-by transistors, such as current mirrors.
The first time I encountered this was when I designed a low-dropout (LDO) voltage regulator in 90nm. It was used as the voltage supply for the rest of my circuitry. At that time, WPE was still a relatively new phenomenon and the transistor models did not take that into account. It wasn’t until when we were very close to tape-out did our modeling team release WPE-enhanced models. However, we had to explicitly turn on this new feature for every single transistor individually in our design if we wanted to see the effects of WPE in our simulations. Unfortunately, for a variety of reasons, some important current mirror transistors inside the LDO did not have WPE turned on. When we got the chip back, the output voltage of one LDO was consistently off by 10% while another one nearby of identical design but slightly different n-well arrangement did not exhibit this behaviour. Another expensive lesson. Now that we’re in 28nm, things have not gotten easier.
Nanoscale side-effects, such as electromigration and WPE, typically only affect analog circuit designers. However, in my next installment on A Matter of Scale, I’ll discuss how digital designers are now starting to feel the pain of scaling as well.
Has shrinking technology affected your own engineering work? If so, I’d love to hear about it.