Page 116 - Analog Circuit Design Art, Science, and Personalities

Milton Wilcox

hastened by the urgency of management, had resulted in mask changes made in
series rather than parallel. And it wasn’t over yet.
Although it might seem obvious that more time invested in evaluation of a new
chip could save substantial time later on, it takes a whole bunch of disciplinc to
continue looking for problems after already identifying two or three that must be
fixed. Somehow we just want to believe that what we have already found is all that
could be wrong. The same advice applies here as to the original circuit design: keep
looking until you can explain every facet of the circuit behavior.
One of the problems discovered in the FET driver was that the theimal detector
was indicating over temperature at 1 1.5 “C instead of the desired 150 “C. I know
that the cause of this 3.5 “C discrepancy had not been discovered, yet one of the
mask changes lowered the reference voltage to bring the shutdown temperature
back up. This again illustrates how vital it is to thoroughly understand the operation
of analog circuits; the silicon had come out with a significant difference from the
design value, and that difference had never been reconciled. If it had been, further
problems might have been averted down the road.
Thc next problem was actually discovered by the niarketiiig manager while he
was rooling around with some of the latest prototypes in his lab at home. He had
noticed that some of the parts were exhibiting a turn-on delay some of the time. He
asked me to check his circuit and I quickly confirmed the problem. When a turn-on
edge was applied to the input, the output would start to charge the power FET gate,
only to latch low again .for several niicroseconds before finally charging the gate
completely. I quickly discovered that the delay was very temperature sensitive; as
the die temperature approached thermal shutdown, the delay becamc hundreds of
microseconds. How could this have been missed? Was it just showing up due to the
mask changes, or as a result of a process problem‘? But most important, exactlj why
wus lhe circuit exhibiting this behavior.?
For anything more than very simple problems, the only cffective way to trouble-
shoot an integrated circuit is by probing the metal interconnections on the chip. As
any IC designer who has spent countless hours peering through a microscope on the
probe station will tell you, this is a long and tedious process. But it’s the only way.
After several days of probing, I kept coming back to the diodes. When an apparently
clean current pulse was applied to thc diode string at turn-on, the diode voltage
initially rose with the current edge, then rapidly collapsed, followed by a slower
recovery to the steady state voltage. It was during this collapse that the thermal
detector momentarily signaled over temperature, thus inhibiting gate turn-on.
Many more days and many more experiments ruled out parasitic resistance.
capacitance, or inductance effects and likewise ruled out coupling via parasitic semi-
conductor structures. Finally, I realized that the reason the diodes were exhibiting
such bizarre behavior n‘as that they weren’t diodes at all. Back when the mask
designer had changed the diode geometries to save area, he or she had unknowingly
created common-collector pnp transistors connected in a triple Darlington configu-
ration! Once again the circuit had becn behaving exactly as it should; it’s just that
up 1.0 this point I had failed to correctly identify the circuit (see Figure 12-2).
Now the pieces of this analog jigsaw puzzle started to fall neatly into place. The
original shutdown temperature had come out low because the DC voltage on the
“diode” string was low. The voltage was low because only the top “diode“ was
conducting the full bias current. In a Darlington configuration, each succeeding pnp
conducts only I/(p+1) timcs the current of the previous device, with the remaining
crirrent flowing to ground. As a result of ample p in these pnp geometries, the
bottom pnp in the string was conducting very low current indeed! And the transient