Someone remind me to post my experiences with the LTZ1000 in about 10 month. I had a design, few hundred of these were built into some equipments, and my NDA will be over by then.
Reminder, spill the beans.
All right, but no specifics. Dont even ask for them, I consider them the IP of my previous employer.
I give you a story instead.
So I was designing high end battery testing equipment. The largest electric car manufacturer, (back in the day, not anymore) was using our system for testing the chemistry and they wanted much better voltage accuracy. Few PPM region. You need this, because open circuit state of charge calculation is very voltage dependent, and errors accumulate etc,etc. So they asked for an improved precision of our system.
We are talking about a big noisy system, which can supply 50A into each battery, and gets hot, even though it is water cooled. Not really the ideal, quiet, nice environment for voltage references. We narrowed it down to the LTZ1000 and the LM399 for the voltage reference, or integrating something into our system (management said no to this before you could finish a sentence). The LM399 was obsolete at that time, even from Linear, I think they were changing to the lead free process, but dont quote me on that.
So I gathered all the circuit information, reverse engineering and pictures that I could, and designed a circuit. And immediately realized how stupid I am because Pin Configuration: BOTTOM VIEW. DOH! Funny enough, if you solder the LTZ1000 on the bottom side it works.
The final circuit is the one from the datasheet. It is on a small board with SMD components (except the opamp and the LTZ), carefully selected 5 PPM/K (max) thin film resistors (1206), SMD substitutes for the diode and transistor. Onboard voltage regulator, is a 7815, nothing fancy. Just place it to the other side of the PCB. And I didnt use any fancy swastika milling or anything like that. There was some to separate the 2N3904 and the VREG, and the resistors, but that's it. Maybe total of 40mm milling, on a credit card sized board. There was also an EEPROM onboard, to store the calibrated values. It went into a small plastic box (since there are big fans in the device) and it attaches to the base board with two nice gold plated board to board .1 inch header. I also placed the 4 critical resistors in close proximity. Thin film, no magic routing, no magic opamp upgrade, no magic low noise voltage regulator.
It worked out beautifully. We couldnt measure reliably the tempco, it was some 0.3ppm for the assembly. After a year we recorded some 2 ppm drift. (Both measurements were done with freshly calibrated 3458A). Inside the system it also worked as it should, making measurements, that are comparable to the 3458A, only faster. Bear in mind that we had to work in a single range, around 4 volt, since we were measuring Li-ion batteries, and nothing more.
Subsequently it was designed into a product, that was selling in larger volumes. Production ramped up. We had a lengthy burn-in process, I think it was at least 2-3 weeks. So I made a PCB holding and powering 5 of these. The output went into a switch matrix that went into the 3458A. Thinking about using a generic purpose relay card to do this? Well, forget it. Normal relays heat up when on, and the contacts will act like thermocouples. So let's design our own switch card.
Calibration setup looked like this: there was a stack of 30 LTZ1000 boards sitting on the base board. All went into another stack of relay cards. That went into the 3458a. Computer controls everything, python. 1000 NPLC voltage reading from the DMM. That is 20 seconds for one reading. One measurement is not measurement. I think we did some 15-30ish measurement of each card, throw away highs and lows, and average the rest. Confidence had to be high. All in the office, done by an engineer. Just making the test setup is expensive. Custom boards designed, firmware written, only to speed up production.
And then the calibration procedure started with an unskippable 30 minutes wait. Before that, there was an earlier setup, where each board was individually connected by hand. So that was about 16 board/day throughput for the calibration phase. Expensive? Well, 16 of these went into machines with "you cannot afford it" price tags, so no. We had a lot more assembly and building to do on each tester.
And that is my story. It has been a great ride! Maybe once I will go back to design precision analog, I definitely miss it sometimes.