How Does an Adjustable Transistor Size CMOS Driver Work?

Question

Background

The source-terminated reflected wave switching scheme is a familiar one. One just needs to add a source-series terminator, which added together with the driver's output impedance equals the line impedance.

I have, however, recently seen the following passage in Right the First Time Vol.1 [p.156] by Lee Ritchey and John Zasio:

There is no need for a separate series resistor if the driving voltage source impedance can be controlled to be equivalent to the transmission line impedance. The drive transistor size can be adjusted so that it produces a half amplitude transition into the transmission line. The impedance of the ON drive transistor acts as the termination. Figure 36.5 shows a half-series transmission line with a series terminated driver and load.

Questions

I'd like to know:

1. how this adjusting the drive transistor size is accomplished. I'm assuming the author means on-the-fly adjustment, and not manufacturing adjustment.

2. I've seen ASICS on which you can select a drive strength. I'd presume this is the same thing, or isn't it? Here's an excerpt from the datasheet of such an ASIC:

Answer 1

The approach to do this is alluded to in the datasheet snippet you linked, which mentions "fingers", or separate parallel gates in a wide MOSFET that forms the output driver. These fingers connect to parallel sources and drains, so when all of the gates turn on together, the arrangement acts as a transistor whose effective width is the sum of the constituent parts.

I don't have an image of an output driver that I'm allowed to share, but the diagram below shows how this kind of layout can be done (using low-voltage core transistors in a 180 nm process from TSMC). I'd imagine that output driver transistors (as opposed to fast, small core transistors) would be longer, wider, and perhaps have additional protection to protect against latchup and other damage induced by the outside world.

Note that each gate is separate (although I connected them to the same net), sources/drains are shared to save space, and everything's in the same diffusion (with dummies) for better matching. This particular design also shows a cascoded transistor, which might be desired in some cases (e.g. if the actual output drive were current-controlled by running the output transistor segments as if they were segments of a current mirror).

By only driving a selected subset of the gates, a desired width (and hence drive strength) can be achieved.