Starbucks of Berlin: Lower Friedrichstrasse

Today I went to Starbucks to think about transistors. To do that, I biked due west on Urbanstrasse, and then turned north to cross the river. This road goes by some of my favorite postmodern buildings, such as the SPD Haus, center of social democratic party politics in Berlin. Also, it hosts my son's Kendo tournaments, I am told.

Approaching the old wall, and crossing it, we are reminded of the past history of surveillance, and the relationship with Eastern Bloc. The Embassy of the Czech Republic is a wonderful futuristic building from the seventies DDR. Unfortunately, it is gated, and I could not get in to snap some photos in dark walkways beneath it, which apparently featured some Soviet ceramic murals.

Oh well, moving on, and looping around to my destination, a spacious Starbucks on Friedrichstrasse. It is nearing some elections in Germany, and apparently the far right party AfD owns the block around this Starbucks!

Now I am thinking about what it means to be far right, or what it means to be social democrat. Since I am writing today about transistors, I am confronted with the bipolar relationship between PNP and NPN. Perhaps the PNP is the dark lord of right wing politics, because it is so over bearing from its populist core? And the NPN is social because they match and talk to each other, with social distancing intact inside the chip?

Let's dive deeper. Entering the 'bucks, I am greeted with Jazz house music, and order a large mug of filter coffee. I want to write a little about some recent research and practical experience with amplifier chips and their positive flavored substrate. 

Did you know that about half the silicon boules in the world are doped positive (with boron) and the other half (with arsenic) negative? We only use the positive ones for chipmaking, whilst negative wafers go almost exclusively to solar panels. I often think about the relationship between my discrete PNP versus NPN transistors, and they come from the two different flavors, but that is actually a small slice of the market besides chipmaking. 

I gazed at the indoor plants, sipped from my huge mug and checked, and the split is about half and half between chips and solar. I know that this system is optimized for holes and electron carriers in operation, but what if some foundry made a process that used negative boules? That is, the emitters of pnp transistors would have the strongest doping, and the npn transistors would be inferior but have access to the body diode for a vertical emitter follower. The body would be connected to positive, and this rail would be thought of more and more as ground, in a negative supply system. Like the old BOSS pedals.

The transistor structures in a regular silicon chip vary widely for different purposes: gain, current and voltage handling, and speed. In the earliest days, pioneers like Bob Widlar had to basically only use NPN transistors, because they were limited by process expenses. The collectors are the first light arsenical plateaus formed, then a heavier boron inset for the bases, and finally a rich arsenic in a tight core forms the emitters. This forms the best beta transistors, isolated from the wafer by reverse diode. 

The process for isolated npn integrated circuits is thus only three steps: n, p, and n. However, to form a pnp transistor, you first have to isolate an area from the body, which is already doped p. So a whole process step is wasted just to isolate the collectors. Furthermore, the rich arsenic isn't used and beta suffers. Silicon designers had to be really tricky, creating different ring structures for emitter and collector with the one boron step available. Since the base is the whole perimeter area, basically beta is upside down and out of control. They are more lateral than vertical and the results were always imprecise.

Later designers transcended the pnp limitations by adding extra processes and using p-channel field effect devices. However, it is important to note the one powerful form of PNP that was always available: a vertical emitter follower with the body as collector. For example, I sipped some coffee and then I started sketching on graph paper, how an LM3900 Norton amplifier would be doped. You can see my results here. It starts on the left with the NPN mirror and gain transistor. Their emitters stare at us like candy in a current bun. But then the fourth transistor is this powerful, PNP that only has two layers: the third layer is the wafer itself, so it is directly modulating current from the grounded chip!

 

In this chip there are also some PNP structures which are formed in the compromised process I mentioned above, with meta-structures and pseudo field effects. They are for micro-current references however, and one of them is basically a boostrapped diode, so they are not in the strict beta matching regime that the input transistors need to hold to. The output of this chip comes from the emitter of the body-collector PNP, coupled to a normal NPN emitter follower.  

This highly effective power relationship pervades most of the opamp output structures you can see, except for Bob Widlars very first effort, the 702 from 1964. One thinks that Bob was in the trickiest mood when he designed this; maybe he had written off PNP as frivolities, maybe he was inspired by older tube designs such as the Philbrick. In any case, he only used NPN transistors, but on his very next design, he immediately gave in and employed the body follower, his first and most powerful PNP placement.

My source is "IC Op-Amps Through the Ages," by Thomas H. Lee. You can see how I circled the PNP follower on the output stage of the OP-07, in Starbucks' house brew. 

Recently I had some experience which really brought the nature of this body follower home. I was using the TDA2822, low power bridge amplifier for a long time, in solar sounders and my line of Tocante solar touch instruments. Realizing the nature of the ground current loop in all amplifiers helped explain a big problem that I fixed by magic.

The ground on these devices was previously connected through a tenuous trace that wound under the input capacitors. Sure it had enough power available, but did it have the rf magic? Apparently, as time goes on, amplifiers like the TDA2822 use smaller, more efficient processes, making their input structure more sensitive and susceptible to magical interference. The size of the PNP body follower probably stayed the same, but it sinks current directly from the wafer, and that can easily start effecting the sensitive current sources that serve as bias and compensating. Basically, the sweetest Tocantes, the Zenert filtered noise and the Phashi sine waves, sounded like fart.

Actually, in the older version, the NJM2073, by Japan National Radio, you can see that the sink output was not a PNP follower, like Bob's original design. This may be intentional, to avoid having a body follower. In any case, they always sounded better, even in poorly grounded situations. The version input, however, always remained a PNP follower with grounded (body) collector. You can see it in both the NJM2073 and the TDA2822. The ground loop would be inevitable once it came to a body follower on the output. 

One day I cut some traces on the fart machines by mistake, including ground. Reconsidering my ground layout fixed everything, in an audio application that would normally not bother with rf magic. Now I'm drinking coffee and gazing at bustling Berlin, and realizing how tied together these PNP transistors are, and they can really mess up the politics of the chip.

 

If you look at my rough sketch of the TDA2822 silicon doping, you can see a sensitive PNP input transistor, and a powerful PNP output transistor all sharing the same boronic piece of silicon ohmically on their collectors. That is why we need to ground analog well. In fact, grounding is more important than bypassing the positive supply!

Next on Starbucks of Berlin: A Butterworth filter in East Side Mall