At the imput to the ADC on the FPIX0 inner board, the analog signal requires approximately 700 ns to settle (from the time is appears at the FPIX0 output pad). This time must be added to the worst case 200 ns required for the digital information to appear. The following three scope images show a near worst case analog signal (corresponding to ~3500 e- input to a pixel cell).
The signal reaches its full value in ~850ns
At 700ns it is 0.4% less than the value at 850ns
At 500ns it is 1.7% less than the value at 850ns
While I have been testing FPIX0 chips, I (DCC) have made a number of mistakes. I hope that by documenting some of those mistakes here, I can:
1: Remember them.
2: Make it less likely that others will repeat them.
The test injection voltage (V-in) should always be in the range 0-3V. In the picture, V_in goes from ground to -0.2V. The signal is only supposed to be injected into cell (11,0), but every cell in columns 6 and 7 (which have simpler kill & inject logic than the other columns) also gets a signal injected, through a transistor which is supposed to be an open switch, but acts as a resistor for small negative signals, and as a closed switch for large negative signals.
Ideally, V-in should be triangular with a very fast negative-going edge, and a very slow rising "edge." In this case, only the negative-going edge will pass through the coupling capacitor.
I haven't yet found an inexpensive pulser capable of making the ideal sawtooth pattern, but for small V_in, a square wave can be used, because the amplifier response is essentially sign symmetric for small signals.
A square wave should not be used for large V-in, since a large "wrong sign" input causes the amplifier to saturate and then overshoot on its return to zero. The circuit takes most of forever to recover.
I have ordered a Stanford Research Systems DS345 30 MHz function & arbitrary waveform generator ($1595 + $495 for GPIB/RS232 interface) which I think is capable of providing a satisfactory test injection voltage. I was concerned that the DS345 might not be fast enough (it can't make a pulse with a leading edge faster than ~35ns), but as these two pictures show, the amplifier response to a pulse with a 35ns leading edge is almost identical to the amplifier response to a pulse with a 6ns leading edge ... For a 250 mV step w/35ns fall time, the pulse height peak is 314 mV; for a 250 mV step w/6ns fall time, the pulse height peak is 318 mV.