From: Karl Guttag Subject: RE: TMS9918
The simple answer to your question is that it was all about cost and ease of implementation. The Home Computer Group provided a lot of the input on how the color was done on the 9918. It was driven by trying to implement the NTSC type signal.
The Video Color Palette on the 9918 was very limited and hard wired. I don't know how much you know about NTSC signal, but the basics are that there is a color burst at ~3.58MHz. The 9918's input clock was ~10.74MHz or exactly 3 times the color burst reference. In NTSC the intensity is the average DC level over a period of time, the color/hue is given by the phase shift from the burst reference of an AC color signal that is at 3.58MHz only shifted, and the saturation (amount of color) is given by the amplitude of the AC signal at ~3.58MHz.
To get color on the 9918, there was a simple resistor tap network. A very long resistor (made out of a long zigzagged polysilicon resitor)was tapped to give various voltage levels. The 10.74 MHz meant that we could get 6 basic color phases/hues just by running a shift register of the 10.74. This shift register was used to control the signals that controlled the tap point on the resistor. A pair of tap points was select, the further apart the tap point were, the more saturated the color and the average of the two tap points set the intensity/brightness (for gray, the two tap points would be the same).
Between the 6 basic phases, we could "tweak" the color by putting in a slow buffer (putting in smaller transistors into a buffer) after the shift register to slightly delay the tap point selection and thus shift the hue slightly. As I remember it, the Home Computer group picked the colors and I remember the colors being adjusted at least once after the first silicon to adjust a few of the colors.
Figures 10 to 12 in the attached patent show how it worked (and Figures 10 and 11 were pretty much taken directly from the schematics).
On the 9928 and 9929 that provided YCrCb, we did a pretty similar D-to-A technique (resistor with tap points) only there were 3 simpler ones since we only had to pick a point for the Y, Cr, and Cb for any given color. With the resistor we had a pretty wide choice of colors.
This is all pretty primitive compared to the latter color palettes also known as RAMDACs which had RAM based color lookups to a more flexible D to A converter. The RAMDACs was a term used by Brooktree and one of the founders of Brooktree, Jeff Teza worked on the Home Computer (it was his first job out of School. I also helped defined several color palettes at TI.
Hopefully the above plus the attached patent helps. If you have any questions, feel free to ask.