Comparing Reverse-biased Transistors
I recently put together two prototypes with transistors from two different manufacturers. I was surprised they generated different results, even though their data sheets were the same.
When I test a new board, the statistic I look at the most is the ratio of ones and zeros. With the Fairchild MMBT304 I see about 0.5% deviation – that is, 49.5% ones, and 50.5% zeros. With ON Semiconductor’s MMBT304 that bias is more like 2%.
My original whitening algorithm could handle 1% deviation but not 2.5%. I’ve since strengthened the whitening algorithm (I’ll explain that in another post), but will probably go with the Fairchild for production runs.
Note: For these tests all three transistors on the board were of the same kind. It's not clear whether it's the amount of avalanche noise that varies between the manufacturers or the manner in which it is amplified.
-0.0052 deviation with monobit test (~0.5%)
-0.0196 deviation with monobit test (~2%)