Keithley 740 system scanning thermometer teardown

[D Straney]

Here's a piece of equipment from the mid-80's: a scanning multi-channel thermocouple reader.


It's got a couple removable cards at the back, one labeled as the "input card" and another as "card 1".

We'll get to the cards, but let's look inside the main instrument first:

That surprisingly heavy card-holder is getting in the way.  Here what it looks like with that removed:


Overall there's nothing too special in there.  There's a display board on the right (which only has 7-segment drivers as far as I can tell), some digital stuff at the top, some power supplies at the bottom, and isolators in between.  The isolators seem to use transformers instead of the now-more-common optocouplers:

Maybe they needed higher data rates than optocouplers at the time were good for?  Or maybe it was a cost thing?
Either way, you can see LM311 comparators at the bottom here, which almost definitely are for receiving the pulses from the other side.  The channels are labeled (left to right):
* Return (->CPU)
* Clock (<-CPU)
* Data (<-CPU)
* Strobe (<-CPU)
I'm assuming these are used to communicate with the ADC in the (floating) measurement section, while the CPU section is connected to earth gnd for its external interfaces.

Speaking of the CPU section, here's the digital stuff:


From left to right, for the main chips you can see:
* The GPIB connector on the back panel
* DS75160 and DS75161 GPIB transceivers
* TMS9914 GPIB interface
* MC68B50 asynchronous serial interface
* LH5168-10L (probably a RAM, the -10 is likely the access time)
* 2 EEPROMs (probably program code)
* ST EF68B09, a 6800-series processor
* Rockwell R6522, a "versatile interface adapter" which seems to provide GPIO and timers for the CPU
* A bunch of octal D-FFs, some misc. logic, and 14 channels worth of ULN2003 drivers (maybe for LEDs?)

There's also a couple bridge rectifiers, some misc. logic, and a GAL programmable-logic device!  I'm guessing this does AC line synchronization for the integration times:

Finally for the main board, there's also an NEC uPD4990A calender/real-time-clock chip (notice the 32 kHz crystal above it, too) accompanied by a dual D-FF and a mystery chip (ITS6728) that I couldn't find anything about:

The '311 comparator over at the right is for the "return" channel of the comms-isolation transformers we saw earlier.


Now what's in the cards?  "Card 1" slot first: this looks like the thermocouple scanner:


That metal block near the right is where the thermocouple leads are attached; it's probably metal to keep all the terminals at the same temperature for accurate cold-junction compensation.  The black blob with a few leads on the bottom side I'd guess is a temperature sensor, for that cold junction compensation.  To the left of that, the red boxes are a bunch of reed relays which select the different thermocouple inputs.  There's lots of guard traces around them on the top and bottom.

Over at the left, there's some RC circuits which may be for filtering the outputs or driving the relay coils (that could be what the ULN2003s on the main board are for).  There's also an ICL7660 switched-cap negative voltage generator.  No idea what that's for.  You can also see that the blue resistor has been reworked, as there's some flux around the leads and missing soldermask underneath it.




Now for the "input card".  There's been no sign of an ADC so far, so that has to be happening on this card.

There's a bunch of terminals (which are all shorted together, strangely enough), a shield, and a notice about only handling the board by the edges (to prevent contamination and current leakage, I guess).  Under the shield:





There's a bunch of analog at the right, and some digital at the left.  There's also a temperature sensor (an AD590 current-output one) near the terminals:

That almost seems like it's meant for cold-junction compensation too, but the sensor isn't thermally coupled very well to the terminals, and despite there being a ton of terminals, the "high" ones are all bussed together, and the "low" ones are all bussed together too, so I'm not sure what the point of this whole section is.

Anyways, since there's no integrated ADC chip on this board, I think the whole thing has to be an ADC implemented from generic building blocks.  There's not enough comparators to make it a flash ADC (and that would be a bad choice for a non-integrated high-precision one anyways), and there's no DAC chip (or DAC resistor arrays) that I can see for a successive-approximation ADC, so my guess is that it's a dual-slope ADC.  These are popular on test equipment too for discrete implementation.

The basic operation of a dual-slope ADC involves an integrator circuit, and counting out time to hit a threshold (in clock pulses).  The integrator is probably this LT1012 low-bias-current op-amp here:

Notice the guard traces around the inputs to prevent leakage currents.  The accuracy of a dual-slope ADC depends heavily on the integrator (it self-corrects for absolute capacitor and resistor tolerance, but other sources of error affect the result), and any bias current flowing into the op-amp input in an integrator circuit is current which isn't flowing through the resistor and capacitor, and so adds an offset error to the conversion.  You can correct for this offset if it's constant, however, op-amp bias current usually varies heavily with temperature; therefore, it makes sense to get rid of the error completely here by using a low-bias-current op-amp.  The two large round film capacitors to the top-right may be the integrating caps.  There's also a small can labeled TG-175 above the LT1012, which according to the little I could find seems to be a matched JFET pair.  The row of transistors to the right of the LT1012 may select different integrating resistors for different ranges or different slopes.

Here's the digital section, labeled with the function of each chip:

Everything seems to match the dual-slope ADC theory.  There's some analog switches (either ranging, or controlling the conversion), a 3.48 Mhz crystal at right and 12 bits worth of counters (for doing the timing), some misc. logic gates (for sequencing), and 16 bits worth of shift register (for shifting out the result serially).  There's also an X24C44 non-volatile memory at the right; I'm guessing this is used to store calibration constants specific to each input card, like offset and gain correction factors.

Anyways, that's all, so here's a final shot of some op-amps and colorful resistors/diodes:

[MadTux]

As far as I know, about all Keithley DMM/Electrometer/etc. with brown plastic case use discrete component ADCs.
Which makes them more interesting and easier+more fun to repair. Anything more modern usually has all the magic happening inside an ASIC or CPLD/FPGA.