-
Nice teardown
Please do a working review, Agilent measurement has some basic videos, but not much in depth.
Would like to see leakage of a glass body diode in the dark and when illuminated - remembering Bob Pease had something on this 20-odd years ago.
ETA: Keithley works around cable noise by pushing the front end out to the DUT:
http://www.keithley.com/products/dcac/sensitive/highresistance/?mn=6430
I always spelled it shmoo
-
Would like to see leakage of a glass body diode in the dark and when illuminated - remembering Bob Pease had something on this 20-odd years ago.
My video of just that on the Keithley 2450 SMU
-
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.
It is fascinating to see inside this instrument and to see the accuracy they can measure to for dc - but for many semiconductor devices such as FETs in particular the resultant curves are only useful for estimating a dc bias point.
Many devices have charge trapped in deep energy levels (ironically generally associated physically with the surface of the semiconductor) which have time constants much slower than RF so when the device is operated under RF the curves it follows can be completely different from those measured under slow dc conditions.
This is a subject close to my heart because the small company I started with two others joined with another small company to develop a pulsed measurement system to take measurements faster than the time constants.
The difference can be spectacular, on some SiC power devices charge trapped in surface states led to the nicely spaced dc curves (as would be measured on a SMU) collapsing more or less into a single curve as the gate voltage was pinned by the surface charge.
So making super accurate dc measurements of semiconductor devices and then using them to fit a large-signal spice like model or even just to estimate the class A power you might get out can lead to spectacular errors much larger than those of using a cheap bench power supply and pocket DVMs to get the same curves!
The characteristic curves depend on where the device is biased and at the rate at which you then measure them.
This is not true of all devices, some we looked at showed very close agreement between pulsed and dc measurements. There will always be some difference because of thermal effects.
-
Would like to see leakage of a glass body diode in the dark and when illuminated - remembering Bob Pease had something on this 20-odd years ago.
My video of just that on the Keithley 2450 SMU
Excellent. Thanks! -
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.
It is fascinating to see inside this instrument and to see the accuracy they can measure to for dc - but for many semiconductor devices such as FETs in particular the resultant curves are only useful for estimating a dc bias point.
Many devices have charge trapped in deep energy levels (ironically generally associated physically with the surface of the semiconductor) which have time constants much slower than RF so when the device is operated under RF the curves it follows can be completely different from those measured under slow dc conditions.
This is a subject close to my heart because the small company I started with two others joined with another small company to develop a pulsed measurement system to take measurements faster than the time constants.
The difference can be spectacular, on some SiC power devices charge trapped in surface states led to the nicely spaced dc curves (as would be measured on a SMU) collapsing more or less into a single curve as the gate voltage was pinned by the surface charge.
So making super accurate dc measurements of semiconductor devices and then using them to fit a large-signal spice like model or even just to estimate the class A power you might get out can lead to spectacular errors much larger than those of using a cheap bench power supply and pocket DVMs to get the same curves!
The characteristic curves depend on where the device is biased and at the rate at which you then measure them.
This is not true of all devices, some we looked at showed very close agreement between pulsed and dc measurements. There will always be some difference because of thermal effects.
Could you give us some info about your product or product planning. I have interest in special purpose device testing as well as modelling of semiconductor effects which affect our products. Foundries like to surpress this critical informations because only some customers have to work with these and most others will be in fear and finally the effort will give no return. -
Could you give us some info about your product or product planning. I have interest in special purpose device testing as well as modelling of semiconductor effects which affect our products. Foundries like to surpress this critical informations because only some customers have to work with these and most others will be in fear and finally the effort will give no return.
It is ten years since I left the field but from my previous experience things probably haven't changed much!
The instrument we produced was called a DiVA and it allowed you to set a bias point and then measure curves from that bias point by pulsing voltage changes to both the gate and drain (for FETs) Voltages returning to the bias point after each pulse. The shortest pulse was (by todays standards) a fairly slow 100nsecs but we found that this was fast enough to be faster than the time constants of the deep levels for trapped charge as well as avoiding thermal changes (the other reason for curves shifting).
Accent, who bought the technology from our little company and for whom I worked for five years were themselves bought out and in the process I think the DiVA ceased to be produced. Various others were producing rival products (such as Keithley I think) and I'd guess that these are still made.
You could probably use a dual channel waveform generator and a scope and a lot of tedious fiddling around to perform the same sort of measurements though this might limit you to a fairly small voltage range. You have to take some care as timing can be an issue (for instance if you have a power limited device biased at high VDS in an almost off state and you want to pulse down to an on state at low VDS you must be sure not to have the gate pulse lead the drain pulse else you might kill the device).
There are a couple of curves (Figures 4.1 and 4.2) in this Masters Thesis
http://etd.fcla.edu/SF/SFE0000249/CharlesBaylisFinalSubmissionMastersThesis.pdf
The thesis also contains a few references but I just tried the following Microwave Journal article we wrote :
http://www.microwavejournal.com/articles/3414-the-importance-of-the-current-voltage-characteristics-of-fets-hemts-and-bipolar-transistors-in-contemporary-circuit-design
and I see that the figures are all wrong - instead of being our figures they are diagrams of on-wafer rf probes as far as I can see!!
If you have stacks of cash you could use Auriga/Agilent kit :
http://www.home.agilent.com/agilent/editorial.jspx?cc=GB&lc=eng&ckey=1963507&nid=-11143.0&id=1963507
but if you're using a foundry the biggest problem may be getting access to test devices.
If you google pulse I-V measurement system you'll find quite a lot of things come up eg. Maury Microwaves system:
http://www.amcad-engineering.com/IMG/File/Pulsed%20IV_product_feature.pdf
though I see that only goes down to 200 nsecs which is probably a little slow. -
Could you give us some info about your product or product planning. I have interest in special purpose device testing as well as modelling of semiconductor effects which affect our products. Foundries like to surpress this critical informations because only some customers have to work with these and most others will be in fear and finally the effort will give no return.
It is ten years since I left the field but from my previous experience things probably haven't changed much!
The instrument we produced was called a DiVA and it allowed you to set a bias point and then measure curves from that bias point by pulsing voltage changes to both the gate and drain (for FETs) Voltages returning to the bias point after each pulse. The shortest pulse was (by todays standards) a fairly slow 100nsecs but we found that this was fast enough to be faster than the time constants of the deep levels for trapped charge as well as avoiding thermal changes (the other reason for curves shifting).
Accent, who bought the technology from our little company and for whom I worked for five years were themselves bought out and in the process I think the DiVA ceased to be produced. Various others were producing rival products (such as Keithley I think) and I'd guess that these are still made.
You could probably use a dual channel waveform generator and a scope and a lot of tedious fiddling around to perform the same sort of measurements though this might limit you to a fairly small voltage range. You have to take some care as timing can be an issue (for instance if you have a power limited device biased at high VDS in an almost off state and you want to pulse down to an on state at low VDS you must be sure not to have the gate pulse lead the drain pulse else you might kill the device).
though I see that only goes down to 200 nsecs which is probably a little slow.
Thank you for the excellent work.
Yes, 200ns is too slow. The problem is more with small foundry devices instead of big packaged discrete devices because the thermal time constant is in proportion to the square of the dimension of the device.
Instead of using a Bias-T I think it is better to use a interface circuit direct attached to the probe head. The issue to serve various devices is more in the current domain, like gummel-plot or subthreshold. So a high current range input on Source/Emitter and ARB driven Gate/Base and Drain/Collector voltage sources are the most effective. The high current range is implemented by an an OpAmp with a number electronical switched feedback resistors. The frequency compensation and speed is then dependend on the current range but that is the trade which I expected everywhere. This output is then feed to the ADC and the hole stuff is put into script control. I guess that with 12-bit DAC + 12-bit ADC the cycle time could go up to 10ns for the mA ranges. The critical point for the speed vs. current is the passive virtual ground capacitance of the probe head. So using a small separate board direct on the probe head with the OpAmp and MUX-Rs should give the best trade.
A side effect is that also Q(V) could be measured by using the low current input on the gate and driving the ramps on source and drain.
Here the DUT Circuit from AMCAD&Maury
-
Instead of using a Bias-T I think it is better to use a interface circuit direct attached to the probe head.
You're very right about bias-Ts especially ones designed for dc rather than pulsed. We didn't use them with the DiVA. But to fair on the designers of that system, the bias-Ts are to allow large-signal S-parameters I guess (though I can't remember if large signal S-parameters are actually large signal S-parameters or small-signal S-parameters measured at each point within the time of the pulse.) -
The part Dave had mentioned in the video (25') is a Sanyu's reed relay.
SORRY about this reply, my English sucks
-
It looks like this http://www.sanyu-usa.com/highinsulation2-en.html or something from here http://www.sanyu-usa.com/category/products/reed-relays2/high-insulation-wet-en
-
Hi Guys,
I just went through the teardown of the B2912A. Truly impressive device, but the teardown got me thinking of how this SMU actually works. The power supply inside generates + and - 245V, fine. Then it uses two constant current sources, one sourcing towards the +245V and the other to the -245V. The current sources are basically made the same as a constant current electronic load, a mosfet and OPA and a shunt resistor. All fine and dandy, but this is the part that interests me. If you have the two current sources, how can the SMU operate in the constant voltage mode? For example, how can you order it to generate a 12V output, if there is no load connected? Even if you order the top current source to let only 1uA through, the voltage on the output would be (almost) the full 245V. Anyone has an idea how the CV part works?
Ivo -
We have a B2912A in school and it was really nice to see the inside of this instrument.
Thanks
I have been watching your videos for a long time, just registered here today.