Fluke 5790A repair project

[TiN]

Some while ago I received friend's 5790A for repair. Condition were quite sad : unit displaying overload error on any selected input range, with nothing connected at inputs. Quick checks revealed worst case : blown Fluke Thermal Sensor chip...

So as result, at least I can decap it and take some crappy photo to show you folks



I wish I'd be able to do better quality photo. Or maybe somebody want to get a chip for proper microscopy shots.

As we can see Fluke went great lengths to thermally insulate the sensor. Sensor heater-transistor pairs are suspended in air, held only by thick metalization trace only. Traces fanout on larger silicon die to gold wire bonds to main alumina chip substrate. Also this silicon carrier die in turn glued to the smaller alumina patch to further help to equalize mechanical and thermal stress?

Same sensor used in Fluke 5700/5720/5730 and 792A's.

[MisterDiodes]

Can you tell if either element is open or shorted?  What was going on when it broke?  Did it work and one day at power up it failed...or did it fail during use??

If you have an open, check the third pad down on left side.  That's a less than stellar wire bond, or could be shadow - those tend to separate with thermals at power up.  There is also contaminant or zap mark just northwest of the lower element...could be leftovers from decapping or some other defect??  It could be normal but an area to look at.

If you have a wafer prober or wire bonder in your collection you can sometimes bring these back to life by stitching on a new bond wire - if the chip survived decapping.

[TiN]

MisterDiodes, 400 ohm input heater reads infinity.

Unit was likely to be overloaded, but I have no detailed history. Lower element is the input side. Pin one is the corner pad on the other side from the Fluke logo.

I don't have any wafer level stuff, nor we deal with these things at work. And I cracked the ceramic substrate anyway (you can see on the photo), so this is a waste for science. I already bought Fluke 5700A donor board to get the FTS sensor (Fluke did not reply the inquiry, as expected ). But before I put good chip in, I have to fix other results of the overload, or the good chip will repeat the outcome as these are rated only for 700mV - 2VRMS input.

Input range resistors are fine though, at least they give proper attenuation levels before reaching the precision AC amplifiers.

[lukier]

Wow, sad story but very cool chip. I wonder how they make it - the inner cutout in the substrate, so the sensors look like being held just by the metalization layer. Maybe back-milling like back-illuminated CMOS sensors.

[SeanB]

Easy to make, you just back etch the chip after metallisation, and mask off the rest of the silicon you want to keep, and stop when you have clearance and have lost a little of the metal layer. Similar things are done to make your MEMS chips and pressure sensors, though there they often stop when there is a thin silicon diaphragm left.

[lukier]

Easy to make, you just back etch the chip after metallisation, and mask off the rest of the silicon you want to keep, and stop when you have clearance and have lost a little of the metal layer. Similar things are done to make your MEMS chips and pressure sensors, though there they often stop when there is a thin silicon diaphragm left.

I was thinking about etching as well, but my only etching experience is PCB etching so I didn't know that one can mask & etch only the back side, without affecting the front gold bonding pads, metalization and substrate. Thanks.

So, in essence, if this chip would have a higher volume it would be dirt cheap like MEMS accelerometers I guess.

[SeanB]

You photoresist off the entire front, metal and all, and gold plate the back side with a mask, then put a layer of photoresist on and etch where it is not covered. Likely this was the first application of back etching ever, as most IC's before only wanted a solderable back and not a thinned out section. The gold is used afterwards, once you strip all the resist front and rear, to solder to the alumina substrate, which in turn has a gold silkscreened layer top and bottom used to bond it to the main substrate. This IC was made in 1986, when hybrids were still top of the pile in complexity, and with a price to match.

[TiN]

Sneak-peak of some test data I've got from after repair and with 5790A under test..

Contenders are : Green corner: 3458A, Blue corner: Wavetek 4920M and Red corner: Fluke 5790A.
 
Source: HP 3245A-LTZ, signal 1 VAC, 100 Hz.

Since all meters agree, we can estimate tempco of HP 3245A is around +15ppm/K for ACV.

Note the difference in noise performance (dots on graph are single samples, lines are average of 8 for 3458A, average of 2 for AC meters).



Test setup

[Noopy]

If someone is looking for some more pictures of the RMS converter I have posted some here:

https://www.eevblog.com/forum/projects/different-die-pictures/msg3664285/#msg3664285