De-capping & circuit analysis of hybrid modules

[magic]

US patent 4,969,823 filed by Analog Devices in 1988 describes a junction isolated complementary bipolar process with N-doped substrate biased to V+ and P-doped epitaxial layer and wells for NPN transistors.



Possibly another hint that this is an AD chip. But AFAIK by the 1990s all big players in high end analog (AD, LT, TI, NS) had some sort of complementary bipolar processes and I don't know them all and how to tell them apart.

edit
Wait, the image above doesn't show separate wells for NPNs. Either it's some other process or maybe it needs those local connections for some reason. Two metal layers aren't making it easy, but I will see if I can figure out this circuitry and which transistors are which polarity.

edit edit
Actually, nothing stops them from running N isolations between NPNs and cutting that P-epi into fragments, for whatever reason.

[D Straney]

Good luck!  You have more patience than me, the dual overlapping metal layers are such a pain - I started trying to follow the connections on the AD586 but had to give up 

Also while finishing up checks on another hybrid, I just ran across an LTC1050, which also has a positive substrate bias in-circuit.  Maybe there's something to be learned by comparing the LTC1050 with the mystery quad op-amp.

[Conrad Hoffman]

I have a question you guys might have some insight on. When I use something like a DRV8837 bridge motor driver, it has a metal pad in the center used for heat dissipation. This is supposed to be soldered to a ground plane. They never say much about it electrically, but I assume it's the actual metal carrier for the die. What might one expect if it were tied to Vcc or even a voltage below ground? Can the silicon be considered a good insulator? Or would there even be a general rule?

[JohnG]

Unless the documentation says otherwise, it is safest to assume that the die is soldered to the metal pad, i.e. the pad is connected to the substrate. For the majority of parts, this would be connected to the most negative supply for the IC.

This pad connection should really should be specified in the datasheet.

John

[D Straney]

If it is connected to the silicon substrate (safe to assume as JohnG said), then what would happen if you connected it to Vcc (or any voltage more positive than Gnd) is that it would forward-bias some or all of the parasitic diodes between the substrate and the various transistors & wells internally, and do all kinds of wacky shit including almost definitely frying the chip.  You might get away with connecting it to something more negative than gnd, but depends on the breakdown voltages, what your supply voltage is compared to the max ratings, etc.

Anyways, back to hybrids and/or mystery op-amps...

[magic]

Good luck!  You have more patience than me, the dual overlapping metal layers are such a pain - I started trying to follow the connections on the AD586 but had to give up 

It's annoying, but I have done AD587 and AD588 in the past, there are full schematics somewhere on this forum. Not sure if I will manage 100% of your opamp, but so far it looks like the input stage is an H-bridge - two diamond buffers driving two ends of a resistor and their collector currents make the output of the input stage. That probably goes to some current mirrors and a unity gain output stage, like in current feedback amplifiers, but not sure yet. Whatever it is, it was likely advertised as a quad high-speed, high slew rate voltage feedback opamp.

A little unexpected type for a thermocouple amplifier, but it was after the mux so they may have wanted speed to quickly scan through all those channels and the precision microvolt job is already done by those LT1012.

Also while finishing up checks on another hybrid, I just ran across an LTC1050, which also has a positive substrate bias in-circuit.  Maybe there's something to be learned by comparing the LTC1050 with the mystery quad op-amp.

That's CMOS, a different technology. It can be done either way (keywords: "N-well" or "P-well"), though P-well (N substrate biased to V+) seems more common in analog for whatever reasons.

I have a question you guys might have some insight on. When I use something like a DRV8837 bridge motor driver, it has a metal pad in the center used for heat dissipation. This is supposed to be soldered to a ground plane. They never say much about it electrically, but I assume it's the actual metal carrier for the die. What might one expect if it were tied to Vcc or even a voltage below ground? Can the silicon be considered a good insulator? Or would there even be a general rule?

Doped silicon is not an insulator at all and the substrate (the bulk of the die) is often connected internally to GND or one supply rail or another, so soldering it to a different rail will do nothing good. Sometimes the datasheet says that the pad is floating, but even then GND may be a good idea for EMI reasons.

[D Straney]

Ahh CMOS, ok, that's what I get for being lazy and not bothering to read the datasheet.  I'm not good enough at this to tell apart weird-bipolar from CMOS unless there's some super obvious interdigitated gate structures.

Anyways, speaking of the LTC1050, this additional hybrid ended up being unexpectedly simple, so in a 2-for-1 deal, here's...
BI 34090218 mystery module
Outside:

Inside:


The IC at the left is an LTC1050 op-amp - the whole right third is taken up by the compensation capacitor:


The IC at the right is a PMI SW-201 or SW-202 quad analog switch:


The two large rectangles in the middle are resistors, covered with polyimide.  Luckily, the coaxial lighting on the microscope penetrates right through that, and we can get a good look at the resistive traces:

At first I was wondering where the damage on the left edge came from, and then realized it's a rough-looking form of resistance trimming.  If you look carefully, you can see some smaller "bypass" paths in between the numbers on the left edge, some of which get cut to increase the resistance slightly.

There's not a lot of circuitry in there, so tracing it was surprisingly quick:


The op-amp forms an integrator (with the ceramic cap visible on the inside, at the far left), which has many possible inputs on different pins, selectable with the analog switch.  I'm not sure what R5 and R6 are doing (the two large polyimide-covered resistors) with their low values: is this supposed to serve as the low side of an external voltage divider?  A current-to-voltage conversion for an external current source?

Overall, the only thing I can think of is that this looks like part of a multi-slope ADC, where various reference and input voltages are integrated sequentially.  However, it's also possible that there's supposed to be a DC feedback resistor(s?) added externally, such as between pins 3 and 13, and it has a more general-purpose use.  This falls into the category, like some of the TRT hybrids earlier in the thread, where the circuit is application-specific enough to not be an obvious general-purpose building block, but not application-specific enough to be obvious where it's meant to be used.

The op-amp is given an internally-regulated supply voltage that's lower than what the analog switch sees.  Here's what R3 & R4, the series power supply resistors, look like:


This is R1, the input-bias-current-balancing resistor:


...and this is R2, the inverting-input resistor pair:


The last interesting thing I noticed is that the diodes used to clamp the op-amp's inverting input look a bit unusual.  There's two bond pads used but they both connect to the same place: maybe a dual diode, with both connected in parallel here.  Can't recognize a Schottky or other special diode structures off the top of my head, but maybe someone else does?

[Conrad Hoffman]

Is BI Beckman and did that go in a meter or some piece of measurement equipment?

[D Straney]

Good question - I only know them as TT/BI, for resistor arrays and trimmers that I sometimes use in designs.  Looks like you're right and "BI" stands for "Beckman Industrial": https://www.ttelectronics.com/products/passive-components/resistors/heritage/
No idea about end-use, as I got it as NOS.  The metal encapsulation and everything is a bit more than it would need even for high-end test equipment, so my best guess is aerospace/military/heavy-industrial.

Edit:
Wait, here's a datasheet for BI's line of "military-grade hybrid microcircuits".  I guess their own part number is the 165-1766-0, and the "165" is their custom hybrid series: https://www.alldatasheet.com/datasheet-pdf/pdf/850180/BITECH/165.html

[magic]

The quad opamp is interesting. The two layers of metal and general appearance are similar to AD586, but there is no AD logo and I failed to find any match in the 1992 amplifier databook.

Well, I must be going blind, because it's right there in this exact book

[D Straney]

MIL-STD-1553 Transceivers, Part 1: National Hybrid NHI-1544FP
Something that I've encountered a lot in looking at older aerospace electronics is the MIL-STD-1553 serial interface.  This was meant as a standard way for avionics to talk to each other - the military and space equivalent of ARINC 429 on airliners.  The higher-level protocol gets somewhat complex with broadcast vs. point-to-point messages, bus controllers, etc., but the physical layer is straightforward.  It uses bipolar differential signalling on a twisted pair, AC-coupled through a transformer to remove any DC grounding issues (Manchester encoding is used to make the AC-coupled pulses timing-dependent rather than level-dependent).

There are a few hybrid modules from different manufacturers which contain an entire physical-layer transmitter and receiver, minus the transformer - the digital transmitter-inputs and receiver-outputs from these are then connected to controllers which handle the higher-level communications protocol.  The pinout seems to be a de facto standard, so this makes a great opportunity for comparing a couple different functionally-equivalent modules side by side, to see how different engineers went about solving the same problems differently!

The first one we're looking at is from National Hybrid, and was found on this Honeywell Aerospace board:


We can see that it's fairly complicated inside, with a lot of different parts:



The ICs inside are all "generic building block" parts:


74LS00 quad 2-input NAND gate


74LS10 triple 3-input NAND gate


74LS26B quad 2-input NAND gate, open-collector outputs


Op-amp of some kind (2 copies)


LM117 adjustable 1A linear regulator


TL431 2.5V shunt voltage reference


LT119 dual fast comparator (photo from a different module of mine)


...plus a couple copies of trimmed resistors that are interesting-looking, and 2 of the power transistors:



For the circuitry, let's look at the receive side first, as it's the simplest:

There are two power supplies: Vcc is +12-15V, and Vee is -12-15V.

• The differential inputs enter at the left, go through series resistors for current-limiting, and have the differential voltage clamped to roughly +/- 1.5V by D5-D8.
• Next, one op-amp in a classic difference-amplifier configuration (U5) converts the clamped differential input into a single-ended signal.
• This signal is then fed through a 3-pole low-pass filter: first, a single-pole RC (R36 & C10), then a 2-pole Sallen-Key type (R37 onwards).  The Sallen-Key topology requires a voltage buffer, and so here a single discrete transistor (Q11) is used as an emitter follower.  This must be to filter out EMI-induced noise that could otherwise appear as false transitions.
• The DC level shift introduced by this discrete-transistor buffer doesn't matter, as the signal is then AC-coupled through C13, and then fed to a pair of comparators (U6).  These compare against separate positive and negative thresholds for the differential input (remember that the input signal is bipolar), and produce 2 separate digital outputs.  The data rate is 1 Mbps, which translates to 1 µs per bit, and there can be up to 2 transitions in this time, because of the Manchester encoding - so the comparator has to be significantly faster than this to work correctly.  The LT119 is specified at an 80 ns delay.  I'm not sure exactly what D9 & D10 are for, on the comparator inputs, as comparator are normally meant to saturate the input transistors.  Their actual effect depends on the relative values of R40 vs. R42-R50.  The TL431 is used here to generate the negative threshold, and the linear regulator (discussed further in the transmit section) is used to generate the positive threshold.
• Finally, a couple NAND gates are used to gate the two digital outputs based on a separate "Enable" ("Strobe") input.  There's also some inverters here made from NAND gates: the output pins can optionally be wirebonded to their outputs instead.  The reason for this seems to be different behaviors when the MIL-STD-1553 bus is idle (~0V differential).  According to a DDC datasheet, the "Smiths-compatible" transceivers put both (Rx Data) and !(Rx Data) high when there's no input, while the "Harris-compatible" transceivers put both (Rx Data) and !(Rx Data) low in the same situation.  Inverting both digital outputs (by bonding to U7B & U7C outputs) and swapping them is equivalent to changing between these two styles of idle-level outputs.  This makes it easy for National Hybrid to manufacture two models of transceivers, one for each "idle style", which are identical except for where a couple wirebonds are placed.

Next, the transmit section:

You can see here where the LM117 is used, to produce an internal supply of roughly 6.6V, which is used as a (not very accurate) reference point in a few places.

There's a lot going on here, so let's step through section-by-section:

• Input logic: The mess of NAND gates here decodes 3 inputs - (Tx Data In), !(Tx Data In), and Tx Inhibit - and produces 3 outputs.  U1B & U1D produce complementary open-collector outputs that are used to modify analog setpoints downstream, and are never active at the same time.  U1C produces an "enable" output, which either powers on or powers off the output driver.  When both "Tx Data In" inputs are complementary, the output driver is enabled and its analog setpoint is modified accordingly.  When the "Tx Data In" inputs are either both high or both low (not a valid state for complementary inputs), the output driver is powered off.  The output driver is also powered off whenever the "Tx Inhibit" input is high.
• Output voltage setpoint generator: A single op-amp is used here in a simple inverting or non-inverting configuration (depending on how you want to think about it), to generate the setpoint for the output voltage.
• When the digital input data is low, U1B's output is low, and its open-collector output pulls the op-amp's "+" input to ground.  This forms an inverting amplifier, where the equivalent input voltage is Vaa (~ +6.6V), so the op-amp's output goes to a negative voltage.
• When the digital input data is high, U1D's output is low, and its open-collector output pulls the intersection of R25 & R26 to ground.  This forms a non-inverting amplifier, where the equivalent input voltage is a scaled (by R19 & R29) copy of Vaa, so the op-amp's output goes to a positive voltage.
• Filters: The output voltage setpoint goes through a 3-pole low-pass filter here, identical in structure to the 3-pole low-pass in the receive section.  As before, a single-pole RC filter is followed by a 2-pole Sallen-Key topology using a single emitter-follower transistor as the voltage buffer.  MIL-STD-1553 has specific specifications for slew rates of the outputs, which is probably for both EMI (less noise induced in adjacent cabling) and signal integrity (less issues with reflections in cabling stubs) reasons.  This filter likely determines the shape of the output transitions, and so makes the waveforms meet the target specifications.
• Bias switch: When the output driver is supposed to be powered off, U1C's open-collector output is pulled to ground.  This turns off Q8, since the Q7-Q8 connection means that Q8's base needs to be about 2*Vbe above ground to turn on.  When the output driver is supposed to be powered on, R13 pulls up Q8's base, and allows Q8 to both turn on the common-emitter Q9 and common-base Q7 with the same collector/emitter current.  Q9 applies +6.6V to the left side of R17 & R17b, which supplies positive bias to the output driver's feedback circuitry, as we'll see soon.  Q7 turns on Q5, which switches negative bias (Vee) to the output driver as well.
• Voltage amplifier & output driver: This part is the most complicated.  Because of the two differential outputs, there are two identical copies of the output driver - the second one is shown at the bottom-right of the schematic.
     
  • The simplest part is the class-AB power stage, made from Q1 and Q2.  The linear output stage, instead of just switches from Vcc & Vee, is needed here because of the defined transition shaping created by the earlier filter, and possibly smaller voltage levels as well?  Q1 & Q2 have their idle bias current set by the "Vbe multiplier" made from Q3, R5, and R6.
  • On the positive output driver, the voltage at the emitter of Q6 is set by the setpoint voltage.  "Bias1" sources current through R17, and Q6's base is at the voltage setpoint (filter output) + 2 * diode Vf (D3 & D4).  One of these diodes roughly compensates for the Vbe voltage drop in Q10, and the other for the Vbe voltage drop in Q6.  Therefore, Q6 acts like an emitter follower with its emitter voltage set to about the voltage setpoint from the op-amp & filter.
  • Q6's emitter behaves as a constant-voltage "virtual ground" summing point for a voltage amplification loop (like an op-amp's inverting input), formed by R8, R2, R2c, Q4, and R7.  Q6's emitter is fixed at the setpoint voltage, as just described.  R8 sinks current from this summing point, while R2 either sources or sinks a current into the summing point depending on the output voltage (this is just like the feedback resistor in an op-amp gain circuit).  R2c connects to the summing point of the negative output driver, and either sources or sinks current depending on the relative levels of the two summing points.  Any excess current from the sum of these various currents is then conducted through Q6's collector, and drives a current mirror (roughly 57:1, based on the resistor ratios) built from Q4.  The current mirror's output provides base current for Q1 and the Vbe multiplier, but more importantly creates a voltage drop across R7, which therefore sets the output voltage for the class-AB driver.  In this way, the "error current" through Q6 is amplified and used to control the output voltage, creating a negative feedback loop.
  • If the output voltage decreases, to keep the sum of summing-point currents equal to zero, more current must flow through Q6.  This increased Q6 current is mirrored and creates a larger voltage drop across R7, increasing the output voltage again and completing the feedback loop.
  • At this point, it's important to point out that the negative output driver's summing point is fixed at 0V, with D3b.  R2c, linking the two output drivers' summing nodes, is responsible for forcing the complementary output voltages.  When the voltage setpoint (from the filter) increases, the positive driver's summing point voltage follows it upwards as discussed earlier.  This puts a larger voltage across R2c, and sources more current into the negative driver's summing point.  To balance this extra current, the negative driver must decrease its output voltage, to increase the current sunk through its own R2b by an equal amount.  This is what creates the "balanced" output voltages, without directly feeding a copy of the setpoint to the negative output driver.
     
     Here's the equivalent circuit made from op-amps, if that makes it easier to understand:
     
     

Let me know if any parts of the explanation don't make sense.

[magic]

Op-amp of some kind (2 copies)

Looks like Harris complementary bipolar process and the chip is somewhat similar to this one, which I thought was a rebadged HA-2520, curiously still "active" at Renesas after 50 years or so. Yours may be some internally compensated variant of that.

Dunno if it's the cheapest way of receiving a differential digital signal.

[D Straney]

MIL-STD-1553 Transceivers, Part 2: Marconi CT3231-M-FP
(Found on this board)
You can see the identical form factor of these modules, at bottom-left and bottom-right, to the previous one:


However, the some of the differences are obvious upon opening:


Here's a 2nd look at the previous National Hybrid module, to show how much more complicated it is:


The Marconi module only has a single layer of conductors (with occasional wirebond jumpers), while the National Hybrid module has at least 2 layers of conductors.  The dark blue color is an insulating layer that separates the two layers.  So at a minimum, the hybrid module itself is simpler & cheaper to manufacture with the Marconi one, needing fewer process steps.

How do they get away with a design that's so much simpler?  The answer lies in the ICs.  There's only 3 of them - the smallest is a simple, off-the-shelf 74LSR00 quad NAND gate:


...but the other two seem custom.  One is marked "CT11":


...and the other is marked "CT12":


They both also contain the text "MCE".  This likely stands for "Micro Circuit Engineering", a British company which seems to have been mostly active in the 70's-90's (can't find any trace of them now).  The "MCE" name pops up occasionally on various mostly-UK avionics I've seen.

Here's the schematic of the receive section:

The connections from U1's input pins 2 & 3 strongly suggest that there's an op-amp inside in a classic difference amplifier configuration.  R15 & R16 are the voltage divider on the non-inverting input, and R17 & R19 are the input & feedback resistors for the inverting input.  The rest of the connected components (R20-R22, C8-C10) seem likely part of an EMI-and-reflections-removal filter, similar to the Sallen-Key filter on the National Hybrid module.
(I think I guessed the Q7 & Q8 connections wrong; connecting the base and collector together makes more sense than collector and emitter, putting them in an "ideal diode" configuration)

Here's the transmit section:


Except for the biasing details of the linear output drivers, everything outside the ICs in both receive and transmit sections is pretty similar to the National Hybrid part.  You can see that they've rolled all the functions of the many logic gates, op-amps, comparators, etc. and even the output-driver's discrete-transistor feedback loop from the previous module into these custom MCE ICs.

There's obviously some tradeoffs here.  With custom ICs, the hybrid module itself is simpler & more reliable, with fewer wiring layers, fewer components to source/inventory/assemble, and many fewer wirebonds and solder joints (possibly the most failure-prone aspects?).  However, they now have the extra time investment, cost, and inflexibility of having to design and then rely on a single source of custom ICs.  The National Hybrid module could've used equivalent ICs from any number of manufacturers if they had sourcing problems - there are plenty of 74LS-series logic gates, op-amps with similar specs, TL431 equivalents, and similar comparators out there.  If Marconi had problems getting their special ICs from MCE, though, it would take a whole lot of time & money (if not licensing issues too!) to take their custom design to a different IC manufacturer, have it manufactured on a new process, and the specs qualified.

In the end, neither choice is "wrong": both have their own advantages and disadvantages, and make more sense in different contexts.  The same tradeoffs appear, even without custom ICs, when designing on the PCB level - do you (1) choose a special-purpose all-in-one chip that comes from one manufacturer and is irreplaceable, or do you (2) make the functions you need out of somewhat-generic building blocks (op-amps, logic gates, etc.) with only a few parts (like processors or ADCs/DACs) that don't have lots of drop-in replacements?  For portable devices, you're often forced to use choice #1, just to fit size constraints - large companies also can take this approach, as they have more leverage with semiconductor manufacturers (or can buy enough chips for a full lifetime production run, so unexpected discontinuation of parts isn't an issue).  For small-run or one-off R&D projects, long-lifetime designs meant to be repairable/manufacturable for decades, or for small groups or companies that are at the whims of semiconductor manufacturers and distributors, though, the second approach is usually better.  The Great COVID Chip Shortage I'm sure left its mark on many engineers - the experience of watching distributor stock fluctuate wildly and having to change designs multiple times in quick succession pushed me from a slight preference for approach #2, to a "follow approach #2 absolutely whenever possible" style of design.

Anyways, hope this was an interesting look inside.  If more of these pin-compatible MIL-STD-1553 transceivers turn up (from DDC, for example) I'll be taking them apart too to compare.

[David Hess]

Looks like Harris complementary bipolar process and the chip is somewhat similar to this one, which I thought was a rebadged HA-2520, curiously still "active" at Renesas after 50 years or so. Yours may be some internally compensated variant of that.

Dunno if it's the cheapest way of receiving a differential digital signal.

If it was from Harris, then their complementary bipolar process used dielectric isolation, so it would have inherent enhanced radiation resistance.  That by itself might be a good enough reason to use it.

[D Straney]

Burr-Brown PGA100 Programmable Gain Amplifier
Continuing my series of "old Burr-Brown hybrids are easy to find cheaply"...

The outside is hard to photograph due to a weird texture on the ceramic lid, but here's the inside with that ceramic lid removed:


The PGA100 is a programmable-gain amplifier.  While a "variable-gain amplifier" usually has continuously-variable gain through an analog voltage, a "programmable-gain amplifier" usually has its gain set (digitally) in discrete steps.  In this case, there are 8 different gains available from 1 to 128, in multiples of 2.  You can see generally how it works from this datasheet block diagram:

In addition to the programmable gain, the module also includes an input multiplexer, to select between 8 separate inputs.  Between the programmable gain and the multiplexing, this was clearly meant to sit in front of an ADC in digital data acquisition systems.

The R-2R ladder shown in the block diagram is an interesting way to select the internal op-amp's feedback ratio and thereby set its gain.  This kind of resistor ladder is normally seen in DACs to create large division ratios with only 2 resistor values, but is useful here for exactly the same reasons.  They work great for programmable attenuators, too (the essential idea behind a "multiplying DAC"): I've made a 20 Mhz digitally-controllable attenuator before by alternately feeding either the input signal or ground to each input on an R-2R ladder, then taking the output signal from the end.  Here, though, it doesn't need to select every possible ratio out of the 256 available - it only needs to select the powers of 2.  Therefore, instead of using 8x SPDT analog switches to switch each "input" of the ladder between signal and ground, it just grounds all the ladder's "inputs" and just taps off the signal at one of 8 places to get the correct voltage divider ratios.  Setting the gain in powers of 2 is also a great way to span a large gain range for an ADC, without needing an excessive number of bits (or circuitry) to set the gain.  Whenever the gain is set as large as possible without saturating the ADC, it's using at least half the full-scale range of the ADC and losing 1 bit or less of resolution (the MSB).

Anyways, now being familiar with the pretty-simple principles of operation, let's take a look at the dies inside...

Resistor ladder
This is the R-2R ladder, made from laser-trimmed thin-film resistors:

This can be seen at the right-hand side in the overview photo above.  I believe (correct me if I'm wrong) that another benefit of using an R-2R ladder here is that for serious precision, similar-shaped resistors will have better-matching temperature characteristics - and an R-2R ladder can be made out of repeated copies of the exact same resistor! (the "2R" sections are just two copies of the "R" section in series)  You can see all the identical repeated shapes on the die here, at least in the bottom 2/3.  I think the upper part is something different, like the fixed parts of the op-amp's feedback.

Op-amp
PMI (Precision Monolithics Inc.) op-amp of unknown part number:


Input mux
PMI 8:1 analog multiplexer, which selects the input channel - notice the 8 large, identical FETs along the top and bottom:

You can also see the common output trace which touches every FET, runs along the outside, and connects to the pad at the right-hand edge.

Digital latch
This is the 74LS378 6x D-flip-flop, as shown in the block diagram above, which latches both the 3 bits of gain control, and the 3 bits of input selection.


Custom Burr-Brown IC
This is the most interesting IC, as it's not an off-the-shelf part.  You can see the "BB" logo at the left edge, showing it's a custom Burr-Brown part.
I think this is the switch that handles the decoding & multiplexing of the op-amp's feedback signal.

My best guess is that they made this custom to minimize the parasitic capacitance.  Most off-the-shelf analog switches use large-area FETs for low on-resistance, but the tradeoff is large capacitance (both across the switch, and to ground) due to the large area.  Since the PGA100 is supposed to work well into the Mhz range, and the feedback resistances have to be reasonably high to avoid large DC power dissipation, even 10s-of-pF either across the switches or to ground will cause problems - either by making a switch look not as "off" as it should be, or by putting an RC pole in the op-amp's feedback path that makes it unstable.

I think the switches here are at the top-left and bottom-left edges, with the digital decoding happening in the middle-left section.  Compare the size of these against the PMI 8:1 mux shown earlier (the dies are similar sizes) and you'll see how much smaller these are.  Since the signal is being fed through these switches into a high-impedance op-amp input, the on-resistance doesn't actually matter very much, so it makes sense to use much smaller transistors with higher on-resistance and lower parasitic capacitances.  (The practical limit for this is where the on-resistance becomes comparable to the R-2R ladder's source impedance, and then any further reduction in switch size balances the reduced capacitance with an increase in the resistance feeding that capacitance, leaving the parasitic RC pole in very roughly the same place)

Hope this was interesting to see, and let me know if I got anything wrong.

[magic]

The opamp is visually similar to LF355/356/357, it should be possible to track it down in PMI databooks, they probably have die images. PMI had improved second sources of those three types, notably with input bias cancellation. You can see that there is a small dummy JFET near IN+ (bottom left) whose gate current is mirrored into the input pins.

I could swear we have seen it before, probably in your hybrid modules because I don't recall Noopy posting it and it surely wasn't me either.

[D Straney]

Thanks, I see the bias cancellation circuit now that you're talking about, hadn't known exactly what that was.

And yes, good memory!  I searched through my die shots and found that it's also the op-amp inside the Burr-Brown ADC-84KG from earlier in this thread:

Looks like the best guess was OP-15/16/17.

[magic]

As expected, there are die images in databooks. I used the 1986 edition, so a slightly earlier die is shown. 1406 checks out, but revision is V instead of U.

I would guess your old one was OP-16 and this is OP-15 (the difference is in capacitors).

[D Straney]

Burr-Brown DAC80 12-bit Digital-to-Analog Converter
This one has already been covered in depth by Noopy here (auf Deutsch), so I'm just going to give a quick overview in English with updated die shots, based on the datasheet and his analysis (check out his color-coded maps of the hybrid and all the ICs).

It's actually a bit hard to track down an accurate datasheet, as I can only find newer datasheets for a later monolithic version of this part.

Overall, the digital-to-analog conversion is done by 3x identical sets of 4 binary-weighted current sinks, combined together in the proper ratios.  This current is then fed to an op-amp set up in a current-to-voltage configuration to produce an output voltage.  This is explained best (as in Noopy's writeup) by this diagram from the ICL8018 datasheet:

(You can ignore most of the part numbers here; the overall scheme is correct though)
The current sinks are created by a set of bipolar transistors with different emitter resistors, and a common base voltage.  An op-amp adjusts the common base voltage to maintain a reference LSB current, so that all of the output currents are scaled as expected.  Again, from the ICL8018 datasheet:

Here's an ICL8018 in the DAC80: each one has 4 switched current sinks, so there's 3 of these dies for 12 bits total.

You can see how the 4 output transistors on the right-hand side are also scaled in size, from top to bottom (1x, 2x, 4x, 8x).  This keeps the transistors' Vbe and therefore the emitter voltages constant even though the current through each one is very different.  (The 8x-large transistor, carrying an 8x current, will have a similar Vbe to the 1x-size transistor carrying a 1x current)

With this DAC scheme, as with any, the MSB currents need to be significantly more precise (percentage-wise) than the LSB currents.  Otherwise, if changing one of the more-significant bits changes the output by the correct amount but with a many-LSBs error, you don't really have 12-bit resolution!  (DAC accuracy and specifications are a whole topic on their own)  The ICL8019 and ICL8020 function the same way as the ICL8018, but are held to looser tolerances than the ICL8018 (0.1% and 1%) so that they can be used for the less-significant groups of bits.  All 3 of the current-sink dies I saw were visually identical: I think the only differences between the ICL8018/8019/8020 are in production-binning, and the design is the same.

Here's the laser-trimmed resistor array containing the 3 sets of binary-weighted resistors that set the currents, plus one resistor as a reference:


The op-amp that sets all the base voltages is this Linear Technology part.  On Noopy's DAC80 this was a PMI OP01, but they seem to have changed the design slightly here:

It drives the common base connection of all the current sink transistors via this discrete PNP transistor (see Noopy's writeup for more details):


A second trimmed thin-film resistor array contains the current dividers that combine the 3 different current sink outputs at /16 and /256 ratios.

Some resistors here are also used for the reference voltage generation.

The on-chip reference voltage is produced by a buried Zener diode.  Mine looks different than Noopy's:


Finally, the op-amp which converts the currents into an output voltage is the PMI (Precision Monolithics Inc.) OP02:


Again, if you want more detail, go read Noopy's writeup, as it covers everything in a lot more depth.

[magic]

This one looks like LM101A, but has a capacitor, which would make it LM107.
According to the 1986 databook, Linear second sourced those two.

Isn't it 101 in the top left corner, with some metal junk on top of it?

[D Straney]

Yes, good spotting! Didn't see that with the metal "2" covering it.

[iMo]

The IC at the left is an LTC1050 op-amp - the whole right third is taken up by the compensation capacitor:

LTC1050 is a zero drift chopper with two "big" integrated S&H caps (the similar LTC1052 has got two 100nF external S&H capacitors)..
Finally we have the 1050 die shot!

PS: the LTC1050 (it is a "low cost" version) has got smaller values of the on-chip capacitors because it chops at much higher clocks (like 2.5kHz), while the 1052 does at around 300Hz, afaik..

[D Straney]

Ahh ok that explains why the capacitor(s) were so large!

HP 3456A resistor arrays
Here's a quick mostly-visual one to knock off the list while I'm working on some of the more complex reverse-engineering projects (and avoiding the actual work that I should be doing).  These two HP modules came to me as random scrap:

Both are built on fiberglass substrates, and seem to be made for spring connections.

It turns out, from doing a lot of searching long after de-capping them, that both are from the HP 3456A 6.5-digit DMM.  You can see them in the board photos in this article on the 3456A, and in TiN's 3456A restoration page.  The HP Journal from 1981, number 4, has a detailed explanation of the HP 3456A starting on page 23 that I'll be referring to later.

The one with the white ceramic lid has a single resistor network inside:


This turns out to provide precision attenuation on the AC/RMS board.  The HP Journal article mentions the AC attenuator on page 28, and refers to a "fineline thin-film resistor" - there's even a photo (Fig.13 on page 29) that looks similar except for the text.  It mentions that the super-long snaking structure at the top-right of my photo above is a 1MΩ resistor, and that no capacitors are used in the attenuator to flatten the frequency response, so it needs a very small parasitic capacitance across this 1MΩ resistance.  According to the article, it achieves this by putting a shield trace next to the resistor on the substrate - if you look back at the fiberglass carrier for the resistor array, I believe the shield trace is the thick gold one coming from a mounting hole on the right-hand side, and running underneath the die.

The hybrid with the black ceramic lid has two resistor networks inside:



This is part of the 3456A's ADC, which uses an interesting multi-slope method (see the HP Journal article for more specific details).  These particular resistors seem to be the precision resistances used to create the integration slopes from the various switched reference sources - the HP Journal article specifically mentions that the ADC is designed to calibrate out its own error sources, and that the only precision components needed are the voltage reference, and tantalum-nitride resistors for setting the slope currents.

Hope you enjoyed the pretty pictures.  I sure do love me some thin-film resistors under the microscope.

[D Straney]

Teledyne-Philbrick 700799 Analog-to-Digital Converter
This one comes from an avionics board that's nice to look at in its own right.

Lid on:

Lid off:



It looks like a successive-approximation ADC - we have all the elements for one, including...

...a DAC with 14 repeated sections (which may translate to 14 bits):

I can't quite figure out the full topology here, but at least part of it looks like an R-2R ladder with the resistors along the bottom edge.  Notice how at the bottom-left, there are repeated sections, each with 5 identical resistors - the top 3 resistors in each section do seem to be arranged in an R-2R pattern, with the top 2 connected in series (forming the "2R").  The much larger resistors above with laser-trimming markings seem to be something different.  They're actually labeled in the metal layer, from R4₁ to R4₁₂, although there's two resistors labeled R4₁ at the right-hand side.  I don't know whether these are both part of the same channel that's different from the rest, or whether this was a mistake.  The IC bears a "Harris '80" marking in the top-right corner.

...an op-amp, which may be a 741 variant judging by the "741K" text on the top edge:

The input differential pair is at the top-left corner, and the big compensation capacitor is obvious at the center-right.

...a comparator:

The input differential pair is at the bottom-left corner, and the large (relatively-)high-current open-collector output transistor takes up the entire top-right section.  Text printed on the die shows "106" in a few places, but this doesn't seem to be the LM106 comparator, as the circuitry doesn't quite match up to the LM106's internal schematic shown on the datasheet - the 2 pads at the right-hand side are not the "balance" and "strobe" pins, as they only connect to each other.

...a smaller ceramic substrate containing a resistor array, which likely contains the precision feedback & summing resistors for the op-amp & comparator:


...plus, a matched transistor array and a few discrete transistors I'm not bothering to show here as they're not very interesting on their own:


On the digital side, we have...
...2 copies of what looks like a 74L04 6x logic inverter:


...and a mystery IC, which I'm guessing is the successive approximation register:

This is made by AMD, as their logo turns up in the bottom-right.  Everything here looks like TTL logic, as it's all bipolar transistors, and snaking blue resistors.
If we count the number of similar-looking repeated sections which have large U-shaped output-driver transistors next to bond pads, the total is 6 on the left edge + 7 on the right edge + 1 on the bottom edge = 14.  This makes a strong case for the module being a 14-bit ADC, along with the 14 repeated trimmed resistors on the DAC.

[D Straney]

HP/Agilent 1NB7-8348, Oscilloscope ADC & sample memory


This came to me already decapped and with no part number - however, an eBay listing showing an identical-looking part shows "1NB7-8348", and the function also matches the circuitry.  The 1NB7-8348 seems to be used in the 54845 & 54846 oscilloscopes, which have a sampling rate up to 8 Gsps, and 1.5 Ghz analog bandwidth.

Just from the looking at the ICs, we can tell that this holds the analog-to-digital conversion stage for one channel of the oscilloscope.  In the photos above, notice the signal that enters on the right-hand side through a transmission line ("conductor-backed co-planar waveguide" layout, similar to microstrip but with some extra ground on the same layer).  This high-frequency input is terminated through a series resistor and capacitor to ground, and then enters the smaller IC:

This must be the high-speed ADC.  The analog signal enters near the bottom-right corner.  It's hard to see what's happening in the bottom half of this IC, as the metal layers cover a lot of the circuitry especially near the bottom-right, but it looks like there's a variety of components (typical with analog) and some large rectangles that could be capacitors.

One popular way to build high-speed ADCs is with multiple pipelined stages of 2-to-4-bit flash ADCs (the simplest but fastest approach to analog-to-digital).  By only converting a few bits at a time, this takes advantage of the speed of flash ADCs, without using an unrealistic number of comparators for higher bit counts.  For example, a 12-bit flash ADC would require 2^12 = 4096 separate comparators: even if we ignore the huge die area this requires, distributing the input signal evenly to all these comparators and decoding all the digital outputs is a serious problem!  If the 12 bits is pipelined in 4 stages of 3 bits each, though, it only needs 4 x (2^3) = 32 separate comparators, a much more realistic number.  You can see an example of a flash ADC near the upper limit of feasibility right here, with the 8-bit/256-comparator AD9012.

Anyways, where this ties back to this particular part, is that these high-speed pipelined ADCs often use capacitors as seen here, and charge-balancing either for the conversion itself, or for subtracting and scaling voltages.  At a minimum, a sampling capacitor is needed for each stage in the pipeline to "remember" the voltage.

Here's a closer view of the analog section:


The top half of the die looks much more digital, and probably handles decoding/latching/timing.  A uniform set of signals with 18 identical cells & 36 pads (18 bits, with differential signalling?) exits this IC, and connects directly to the larger IC.

The larger IC is very clearly RAM:


I didn't image the middle of this die because it's very repetitive - just an identical set of memory cells repeating thousands of times.

This is the sample memory FIFO: with the high sample rate (8 Gsps), the ADC is pumping out digital data faster than any processor at the time could realistically digest.  Because each waveform capture is a certain length, the easiest thing to do is to dump all the sample data into a very fast dual-port memory that can keep up with the high write speed.  When the waveform capture is done, the rest of the scope can read out this sample data at a more leisurely pace, and take its time processing the data at least until the next trigger happens.

Here's a closer view of some of the logic happening around the edges:


Hope this was an informative look at how high-speed oscilloscopes worked a couple decades ago.