EEVblog #607 - Agilent B2912A Source Measure Unit SMU Teardown

[TiN]

First I thought that's a heater on a fan, to maintain constant temperature inside unit, but since it's exhaust fan, no, just a shield

Surprised to see MiniFit JR connector on output side, and no triax on this SMU, and still spec'd as 10fA!

[Tek_TDS220]

To FrankBuss: Interesting idea to use the 24-bit A/D for software defined radio.  I didn't know they made 24-bit devices that fast.  You'll certainly need a front-end with a high pass filter.  If you don't use a bandpass filter you'll also get a lot of high frequency noise that 'folds-in'.  Keep in mind that the most advanced FlexRadio SDR's use two 16-bit 250 MS/sec A/D's with bandpass filters.

[dentaku]

Wow, talk about NSFW. That's one hell of an instrument. 10fA! Is there a little dude in there somewhere counting the electrons as they pass by? I couldn't see him...

10fA is one single electron per 16µs, if anyone didn't realize that.

I did not... that's amazing

[hans]

I really wonder what kind of thermal performance they can get out of those TO263's. If it's all linear (high precision, low noise) device, and can deliver up to 30W, and need a fan ducted heatsink, I can imagine they can probably dissipate 30W on the PCB as well?
edit: it can deliver 105mA up to 21-210V. So at 22V out, and assuming 245V input, then (245-22)*0.105 = 23W.

I'm very tempted to back-mount a TO220 heatsink (with 1 flat side) and see how much thermal resistance the PCB "adds", and via-stitch the hell out of the copper planes.
If I'd use a 5K/W heatsink, how much would the overall resistance become? 10K/W? Less/more?
Even with 10K/W, you can probably dissipate 6-8W surface-mount at room temperature. Maybe even more with a fan directly on the heatsink..
Would be interesting to try for some really compact cooling..

[tx8]

Is it good practice to wind the long wires up like that? Sure looks like candy sticks but I heard of computer systems malfunctioning because they wound data bus wires around cassis parts or something like that causing inductance and magnetic interferances. 



PS.: Or doesnt it matter as long as its only a couple not winding around a center piece of metal?


PPS: Got it.:  http://en.wikipedia.org/wiki/Twisted_pair Sorry this post has become obsolete.

[retrolefty]

Dave mentioned that one might find a rack mounted version of this on a manufacture assembly run. I can see this used in a companies R&D lab, but what use would it have in unit assembly/testing? Seems like it's main use is characterizing individual components? Other applications would be?

[Christe4nM]

Awesome bit of gear! Thanks for the elaborate teardown Dave! And yes I would certainly want to see it in use.

I can't help but pausing the video many times and just stare at the electronics and even mechanical stuff that just shouts quality engineering through and through, trying to take it in. Would love to be part of a team designing such an instrument one day. (A man can dream, right )

As for semiconductor testing during manufacturing that is the domain of big test machines with custom fixtures for each type of IC. I've seen quite a few different ones at NXP Nijmegen where they test the testers and test algorithms before they go life on the actual testers in their manufacturing plants. On the other hand those testers are basically 'just' a rack of 4 quadrant powersupplies/SMUs PCBs on a backplane connected up to the actual test head.

[kilobyte]

The U1103 in minute 26 could be a ADG1212YCPZ the datasheet says that the branding is S08

[German_EE]

That 24-bit ADC would make an amazing multimeter front end. With a span of +/- 2V you could have a quarter of a microvolt resolution.

[robrenz]

That 24-bit ADC would make an amazing multimeter front end. With a span of +/- 2V you could have a quarter of a microvolt resolution.

The Keithley 2450 has 10pV resolution on the saved data on the lowest range. Displayed measure resolution is 10nV. But it is not as accurate as my Fluke 8846A as seen in my 2450 SMU review here

[FrankBuss]

To FrankBuss: Interesting idea to use the 24-bit A/D for software defined radio.  I didn't know they made 24-bit devices that fast.  You'll certainly need a front-end with a high pass filter.  If you don't use a bandpass filter you'll also get a lot of high frequency noise that 'folds-in'.  Keep in mind that the most advanced FlexRadio SDR's use two 16-bit 250 MS/sec A/D's with bandpass filters.

Good idea to use a bandpass filter. Maybe I should use 100 kHz as the lower limit, for filtering most noise like mains, sferic (on the other hand might be interesting to measure sferic) etc., and the upper limit well below the nyquist limit. Then it should be possible to do the rest digitally.

[plesa]

I added some pictures to compare the Agilent B2912A SMU with Keithley 2636B SMU. It has similar construction with heatsink tunnel in center of instrument shared with two identical modules.
For obvious reasons I do not want to dissassemble it fully.
The power supply is Densei-Lambda.
When I was looking for new SMU the Agilent was also one candidate.The spec are quite simmilar.
The controlling instrument by LXI or by embeded TSP is difficult, I rather control instrument by LabView.

[plesa]

more  photos

[chickenHeadKnob]

Plesa, those are interesting pictures. Are the vertically oriented boards all digital system? It would seem they would suffer from a large temperature gradient top to bottom.

[hikariuk]

Dave mentioned that one might find a rack mounted version of this on a manufacture assembly run. I can see this used in a companies R&D lab, but what use would it have in unit assembly/testing? Seems like it's main use is characterizing individual components? Other applications would be?

Verifying that randomly selected components from a production run match the reference characterization?

[plesa]

There is only one board oriented vertically, the rest is oriented horizontal. The SMU boards are oriented both way and on bottom side of instrument is power supply, digital board and one board which is connected between Densei-Lambda power supply and SMU analog boards.
The instrumet fan is almost silent after few minutes when it is powered up it became quite noisy.
I can make some Flir images if you want.

added front end pictures.
I was quite surprissed with huge oldschool 4 terminal power resistor in front end enclosure VCS302 0R1.On front end is quite popular LMC66.

[GregRexUzelac]

Beautiful! Thank you, Dave.

[KC0PPH]

First post here..

First of all Dave, Have watched a fair amount of your videos... upset you never finished your Power Supply...

Anyways On Topic... Since he never finished his lab supply (and I am needing one) I decided to build one (I am an Industrial guy so dont get around "professionally" a lot of the more traditional electronics stuff a lot), I am still in the design phase and figured out this SMU is the same thing that im trying to build (goal was 1uA not 10fA though).

I learned a lot from this video and will incorporate their design practices into my design (who knows might make it OH or put it on Kickstarter)

So has anyone heard of a "Service Manual" for this guy? (Doubt it lol) ...   

[rf-design]

Hi Dave,

I could not recommend this SMU for chiplevel semiconductor device measurements and I think for general purpose SMU jobs in the board level domain it is too expensive.

The panel show a 10fA number but that is only the resolution. At 10nA range the offset is 50pA or 5000ppm. That is because of the hole design compromize where the high current range MOSFETs present leakage to the output path. That is because I guess the relays are not used to isolate the current ranges. I had issues in the past where the temperature rise in high current range change the gate leakage current impact in a subsequent low current range. That could be fixed by reversing the sequence in testing but makes the device stepping slower. I tought that it would be better to use relays there. Optional these could be inactive by software if speed is prefered.

The second issue is that todays small semiconductors have temperature time constants which are 10-100x faster than 10us. So the curves are distorted by the various temperature effects. These are important only in the higher current range relative to the maximum current of the devices but there is no trade built into the SMU instrument speed versus resolution.

So for semiconductor SMU jobs it is not specific designed for and for discrete device testing it is too expensive.

[coreman]

Hi Dave,
 I enjoy your videos thanks.  Those triaxial connectors that are used on the external test enclosure connections are the same that are used on a jet aircraft data distribution bus, ie MIL-STD-1553.  i have worked on those in the past. the coax used with them was 78 Ohm.  After watching your video i then realized that maybe they were using a similar "guard" screen system on the aircraft.  Aircraft power was 400Hz with the chassis the return.  but i think the 1553 data bus is a differential system.  anyway that is first time i have seen those connectors any where else other than on aircraft.  they could be used on land vehicles, space craft etc too...

[plesa]

Hi Dave,

I could not recommend this SMU for chiplevel semiconductor device measurements and I think for general purpose SMU jobs in the board level domain it is too expensive.

The panel show a 10fA number but that is only the resolution. At 10nA range the offset is 50pA or 5000ppm. That is because of the hole design compromize where the high current range MOSFETs present leakage to the output path. That is because I guess the relays are not used to isolate the current ranges. I had issues in the past where the temperature rise in high current range change the gate leakage current impact in a subsequent low current range. That could be fixed by reversing the sequence in testing but makes the device stepping slower. I tought that it would be better to use relays there. Optional these could be inactive by software if speed is prefered.

The second issue is that todays small semiconductors have temperature time constants which are 10-100x faster than 10us. So the curves are distorted by the various temperature effects. These are important only in the higher current range relative to the maximum current of the devices but there is no trade built into the SMU instrument speed versus resolution.

So for semiconductor SMU jobs it is not specific designed for and for discrete device testing it is too expensive.

I have the same opinion, that it is too expensive. I  do not thing that banana socket is best chioce for the precision measurement. On 2636 it is much more better, lowest range 100pA. For basic  semiconductor characterization is better choice the Keithley picoammeter 6485 which is about 1,5k USD but without supply. Or their dual picoammeter 6482 with integrated supply (30V ) for about 4,5k USD.

Hi Dave,
 I enjoy your videos thanks.  Those triaxial connectors that are used on the external test enclosure connections are the same that are used on a jet aircraft data distribution bus, ie MIL-STD-1553.  i have worked on those in the past. the coax used with them was 78 Ohm.  After watching your video i then realized that maybe they were using a similar "guard" screen system on the aircraft.  Aircraft power was 400Hz with the chassis the return.  but i think the 1553 data bus is a differential system.  anyway that is first time i have seen those connectors any where else other than on aircraft.  they could be used on land vehicles, space craft etc too...

There is much more triaxial connectors available - not only BNC basis (Fischer, Lemo they have their own solution), the old ones has only two lugs and can be destroyed plugging it to normal BNC, thats why has been made the tri-lug triaxial connector.
For purposes you mentioned were propably used connectors with different insulation materials (cheaper).
I was also quite surprised when I saw the triax during Mike teardown of the Fluorescence Spectrometer. For low level measuremnts should be used only connectors with PTFE insulation.
It is quite similar with cables, you can find the cheap triaxial cable, but it is not useable for low level measuremets in pA range.

[rf-design]

For comparison the Keithley 617 have

10fA Error at the 200pA Range

instead of

50pA Error at the 10nA range

That shows that the choosen SMU architecture is limited and the bad thing is that is more a matter of some (# ranges) relays.

There was some hint that it is not designed by agilent but bought. Is there a link?

[rf-design]

There is much more triaxial connectors available - not only BNC basis (Fischer, Lemo they have their own solution), the old ones has only two lugs and can be destroyed plugging it to normal BNC, thats why has been made the tri-lug triaxial connector.
For purposes you mentioned were propably used connectors with different insulation materials (cheaper).
I was also quite surprised when I saw the triax during Mike teardown of the Fluorescence Spectrometer. For low level measuremnts should be used only connectors with PTFE insulation.
It is quite similar with cables, you can find the cheap triaxial cable, but it is not useable for low level measuremets in pA range.

Here is a hint from the 617 manual about the cable material to minimize current noise in the fA range.

[Zucca]

Just to give the full picture, in a Triax the middle is not always the guard as described by David here

https://www.youtube.com/watch?feature=player_detailpage&v=pKX50E_14MQ#t=3034

In my Keithley 220 the outside is the chassis, the middle is the ground and the center is the output.

See here for details:

https://www.eevblog.com/forum/testgear/keithley-220-output-triax-connector-mod/msg397850/#msg397850

the guard is provided in the back with a banana female, separated from the Triax cable.

I don´t see why the Keithley 220 engineers decided to do that. Maybe the chassis can not always be used as a return conductor for a tiny current.

[plesa]

There is much more triaxial connectors available - not only BNC basis (Fischer, Lemo they have their own solution), the old ones has only two lugs and can be destroyed plugging it to normal BNC, thats why has been made the tri-lug triaxial connector.
For purposes you mentioned were propably used connectors with different insulation materials (cheaper).
I was also quite surprised when I saw the triax during Mike teardown of the Fluorescence Spectrometer. For low level measuremnts should be used only connectors with PTFE insulation.
It is quite similar with cables, you can find the cheap triaxial cable, but it is not useable for low level measuremets in pA range.

Here is a hint from the 617 manual about the cable material to minimize current noise in the fA range.

http://www.keithley.com/knowledgecenter/knowledgecenter_pdf/LowLevMsHandbk.pdf