Keithley 2015 teardown

[muvideo]

Hello, let's see whats inside a keithley 2015THD

It's a generic multimeter that can measure also THD of AC signals in audio frequency, it contains
a sine wave generator with variable voltage and frequency output. The frequency
is audio 10Hz-20kHz, the voltage goes up to 4V rms.
I received it a little shaken up by post, I feared of big damage when I saw that the front panel
and display weren't in position anymore. Fortunately it was all ok.




Let's open it:
Mainboard, this is all the standard multimeter:


The black vertical box with metallic tab is the 100mOhm current shunt,
the PLCC IC soldered on the board between the crystal and the rear connector
is the main ADC for basic multimeter functions:


This is the voltage reference, I was sure it was the LM399, but it's marked LT,
has 4 pins under it, dont know what it is.


Digital control, processor and firmware, divided in even and odd bytes:

[muvideo]

Seem that keithley shares the board with a normal 2000 multimeter:


Another view of the mainboard, the signals come from low left,
where there is protection circuitry, goes to input shunts and scaling, AC top left,
and goes finally to ADC low right, from there to the digital section trough the
optocouplers


here in transparency, the 4 layer board shows the two voltage domains:


In the 2000 multimeter, under the mainboard there is the space used for
plugin scanner cards, in the 2015 this space is occupied by the THD board:


This board contains many interesting things, one dsp, tree fpga, two dds and one adc,
plus a other components and optocouplers. There are tree isolated domains:
in the middle left there is the analogue input signal, from analogue board input,
that goes to a dedicated adc AD7722, the clock here is generated by the first AD9850.
from here the signal crosses the four upper optocouplers and goes to the dsp where the
THD data is extracted and sent to the digital board using the connector in up left corner.
The DSP controls also the sinewave generation, using the tree lower optocouplers,
communicating with the lower fpga and to the second AD9850, from here the signal
is filtered and amplified, low center to the two bnc rear connectors.

[muvideo]

The meter works ok, it is full of dust internally, I have only to decide if I will wash it,
I'm tempted but there are all that bug sad (I meant big smd, my keyboard has autonomous thinking    )
components... but I dont like all that dust, probably I will use some distilled water and a little heating to dry the board.

Fabio.

Edit, another pic, display board:

[mikeselectricstuff]

I'd guess that reference is an insulated assembly containing a heater with temperature control to hold the ref at  a constant temp. 

[alm]

That LT voltage reference may well be an LM-x99, LT is a second source for this chip. If you're going to wash it, be very careful about leaving residue. The input impedance is probably specified as > 10 Gohm. It does not take a lot of leakage to fail this spec.

[muvideo]

That LT voltage reference may well be an LM-x99, LT is a second source for this chip.

I've heard that the 2000 uses the LM399 (or 299?), so it's likely as you said.
I'll measure the voltage on it, if it's an LMx99 equivalent I'll find 6.95V on the zener.

If you're going to wash it, be very careful about leaving residue. The input impedance is probably specified as > 10 Gohm. It does not take a lot of leakage to fail this spec.

I used the 7061 to measure the 2015 input resistance directly and using a 10Mohm resistor in series with the input,
it reported 10Gohm of input resistance at 6,5V test voltage, while the 2015 was in DCV 10V range.
The input leackage measured  with 10Mohm sort of confirmed this.
The other (older)  meteres I have seem to have lower input leackage.

I dont have access to IPA, so probably distilled water can be good enough, I washed the 7075 with water
and soap, it was very dirty and greasy, but it had some leackage problems in the first hours of work after washing.
After a pair of weeks of work it seem to have regained decently low leackage, but I had not much
to lose with it

Fabio.

[metalphreak]

Taking transparency photographs of PCBs should be a requirement of all teardowns in the future

[muvideo]

I risked and washed the boards, first pass water and soap (a degreaser that I had tested and
I know it doesnt leave visible residue back), from the board dropped black water!
Second pass with water only, waited a little and third pass rinsed with demineralized water.
I let the board drip off most of the water, caring that the analogue part was on top,
a little compressed air and the boards were like new. I protected the piezo beeper
with a little tape and tried to let go as little water as possible to the input selector switch.

I tested the meter for leackage and it seem still ok, measured the same or a little more than before.
I noticed that the input current varies considerably with the acquisition speed and voltage range,
apart the 10Mohm of 100 and 1000V ranges.
When I'll have a tek 7A22 amplifier I will play a little to see what's happanening to the meters input

I also measured the reference and it is slightly less than 7 Volts.

Taking transparency photographs of PCBs should be a requirement of all teardowns in the future

Glad you like it, I've rarely seen so transparent boards, copper areas contains very little copper
and the traces are very thin, you can easily see trough ground planes.
Seem reasonably serviceable, there are only few special parts like the main adc and the voltage reference,
it's a shame that keithley doesnt release schematics, at least to the public.

Fabio.

[M. András]

what was its original price? nice piece of equipment. the traces on the digital parts are truely fine and dense wish i could get nice deals here in hungary, but unfonrtunatly most of the good stuffs are in the usa and the shipping+custom prices are huge

[muvideo]

what was its original price? nice piece of equipment. the traces on the digital parts are truely fine and dense wish i could get nice deals here in hungary, but unfonrtunatly most of the good stuffs are in the usa and the shipping+custom prices are huge

I dont know the price when new, but I wouldnt be surprised if it would cost much more than a regular 2000.
It should be still current in keithley line, but it is around from many years (nineties?),
so it's easy to find on the second hand market. It did cost to me not too much, the seller accepted to
sell at a lower price than advertised (mainly luck and the right offer at the right time),
I was on the lookout for the right combination from some time.
I'm also in eu, I know that it hurts when you see the equipment cost be lower than delivery cost,
but on rare occasion there are true bargains, also considering the other costs.


Fabio.

[free_electron]

It is an lm199. Linear made them under the same name. They are obsolete , unless your name is agilent or keithley or fluke and order a lot of wafers once in a while...

Very expensive component...

Since you got the keithley dmm. IMMEDIATELY change all the caps in the power supply. I have seen at  least 5 keithley multimeters 2001 / 2002 with shorted out electrolytics. It fries the motherboard before the fuse pops.... I salvaged the lm199's so i got a few of those laying around.

Bob pease has a nice video on youtube where he uses 10 of these lm199's to make a super stable reference..... Itll cost you 500$ in 'zeners'....

[muvideo]

Thank you for the warning, since you opened them, do the 2001/2002 share the same voltage reference
with the "economic" 2000, if not what reference do they use ?  There is not much info around.

Fabio.

[codeboy2k]

It is an lm199. Linear made them under the same name. They are obsolete , unless your name is agilent or keithley or fluke and order a lot of wafers once in a while...

Very expensive component...

Obsolete is right. LM199/299/399 is overrated now, and praised far too much. It did the job well at the time it was created,  but there are far
better references now, at decent prices < $9 US and no heater.   Do the big 3 above still use LM199's in their new designs??

Here's a sampling of some good references compared to the LM199/299/399 series:

part	Voltage	tempco	initial accuracy	RMS noise
LM199/299/399	6.95	1 ppm/°C	2%	7 uV
MAX6350	5.0	1 ppm/°C	0.02%	3.0 uV
LTC6655-5.0	5.0	2 ppm/°C	0.025%	1.25 uV
LTC6655-2.5	2.5	2 ppm/°C	0.025%	625 nV

Bob pease has a nice video on youtube where he uses 10 of these lm199's to make a super stable reference..... Itll cost you 500$ in 'zeners'....

I'll have to hunt that down, so I can make a super stable reference using 10 MAX6350's at only $90 and an order of magnitude less current

[alm]

LM199 has a TC of 0.2 ppm/K typical. Manufacturers like Keithley use selected references, so the typical spec may be their max. Long term stability at 1000 hours is specified as 20 ppm typ, and can be much better after burn in and selection.

Agilent and Keithley have a lot of data on aging of these references. The reference used in Agilent's reference multimeter and several voltage standards is the LTZ1000, another heated zener. I think its tempco is even lower than the LM199. Heating is an excellent way to keep the TC down if power is not an issue, which is why we still use OCXOs.

[muvideo]

Obsolete is right. LM199/299/399 is overrated now, and praised far too much. It did the job well at the time it was created,  but there are far
better references now, at decent prices < $9 US and no heater.   Do the big 3 above still use LM199's in their new designs??

cut...

I'll have to hunt that down, so I can make a super stable reference using 10 MAX6350's at only $90 and an order of magnitude less current

You should be horrified then, checking the reference used back then in the solartron/datron meters,
that use discrete zener(s), with some selection and compensations
Recently someone on volt-nuts list posted the factory calibration procedure at solartron for 7081,
the work done selecting and aging the parts is big, but the final specs are high, considering the
"cheap" parts and design used, I think most of the cost was in the hours of work needed to finalize every
single meter: http://www.febo.com/pipermail/volt-nuts/2012-May/001860.html

As for the list you posted, I think there is more than that specs, the initial accuracy being non issue,
but here I'm no expert, i found useful info in some LT's app notes, for example AN82 and AN86
(this one per Jim Williams tradition contains all the juice in the appendixes   )

Fabio.

[codeboy2k]

LM199 has a TC of 0.2 ppm/K typical. Manufacturers like Keithley use selected references, so the typical spec may be their max. Long term stability at 1000 hours is specified as 20 ppm typ, and can be much better after burn in and selection.

Agilent and Keithley have a lot of data on aging of these references. The reference used in Agilent's reference multimeter and several voltage standards is the LTZ1000, another heated zener. I think its tempco is even lower than the LM199. Heating is an excellent way to keep the TC down if power is not an issue, which is why we still use OCXOs.

Perhaps that's the real key, that they have such confidence in these older references due to the aging data they've collected; If I wanted a super stable reference using one of the newer reference IC's, I think I too would probably put it in a temperature controlled oven. It should obviously be at a stable temperature during operation. I'm aware of the LTZ1000. It's really expensive 

Has anyone ever done this before, i.e., built a modern reference IC into a home-brew oven. Obviously, if you do, you first have to have an actual need to to that.

You  could build a small board with a reference IC, and a couple 10 ohm, 1/2W resistors in series placed on either side of the chip, raised into the air to heat the surrounding air, not the board or the chip directly.  The board would includes a temperature sensor and an opamp controller.  Then box it up, insulate it , etc, etc. Power with 3.3 - 5.0V; the heater circuit would consume about 160 - 200 ma, and produce about 500mw - 1W of heat. The temperature goal would be about 50°C , so it can still work in the hottest parts of the world.  I haven't done the math, yet... oh, dammit... let's do the math here...

assume the oven space is 1 cubic inch (1.64e-5 m^3 or 16.4 cc)
assume the air is dry
assume we want to work at ambient -10°C up to +40°C
      (ambient might need to be even higher if it's placed in an environment already producing heat)
assume the oven temperature we want to achieve is 50°C (10°C above maximum expected ambient)

Therefore, at the far end of the extreme, we need enough power to raise 16.4cc of air by 60°C.

According to Wolfram Alpha , 16.4cc of dry air weighs 20.9 mg (0.0209 grams)

The specific heat capacity of dry air is 1.007 Joules / g°C difference.
.0209 g  * 1.007 J/g°C * 60°C = 1.2627 Joules are needed. 
1 J = 1Ws, so 1.26 Watts for 1 second.  or 500mW for 2.52 seconds. or 250mW for approx. 5 seconds... etc.

So I think 500mW to 1W of heating capacity would work fine in this oven. Maybe even too much, you could even reduce it. 
250mW would take a few seconds at start-up to warm the air, but after that it would barely need much more  to keep it warm.
I'd probably finally end up choosing 1/2W resistors dissipating 100mW each and let the controller do it's thing.
Maybe only need 1 resistor and 50mW dissipation to save power. It would take about 30 seconds to 1 minute probably. 
Now you can't go too low in power, because at -10°C you're losing heat to the ambient very quickly, even with insulation,
and you have to overcome those losses too. Sigh.. so much math, too little time

I might build this up one day, so comments on my thought process and calculations are appreciated..

[alm]

Yes, putting a normal reference in a DIY oven is definitely feasible, I believe some people on the volt nuts mailing list have done it. A higher temperature is easier to keep constant (oven temperature close to ambient gets really hard to control), but increases the aging (drift) rate, so something like 50°C would probably be better than the 80°C or so used by HP et al. because their stuff has to work across the full commercial temp range. Once you have decent control, I would expect your biggest enemy to be temperature gradients. Temperature gradients induce thermocouple effects. I believe the LTZ1000 datasheet talks about keeping both tracks to the reference the same length and width, so the temperature gradient is the same on both sides (metal junctions at same temperature means thermocouple effects cancel).

[muvideo]

Has anyone ever done this before, i.e., built a modern reference IC into a home-brew oven. Obviously, if you do, you first have to have an actual need to to that.
...

I might build this up one day, so comments on my thought process and calculations are appreciated..

I think it's not hard to build an oven, at that size the heat capacity of entrapped air is very little, and a transient
problem only, will be much more the heat capacity of solid parts in the oven, probably you want the oven
controller in the oven itself so you can stabilize also the controller, if you want rapid heating you want good
controller feedback stability to avoid temperature overshoot.
Edit: naturally after the transient the power usage depends on the thermal conductivity of the walls and,
being this not zero, depends also on the oven outer surface extension.
If you can obtain good voltage stability
probably you will not power down the thing to not upset the equilibrium obtained.
As for voltage stability it's not easy, there are many subtle things going, I was warned in the past on
this same forum (thermoucouple effects, aging, humidity ).

There is much work already done on the net about voltage references, on volt-nuts, and there are
hardcore builders in some chinese forums, they give a lot of useful info but it's a little hard to read
http://www.ourdev.cn/forum.php?mod=viewthread&tid=3593996
or bbs.38hot.net, for example this is a ltz1000 based reference pcb with thermal oven cutouts
and strain reliefs:
http://bbs.38hot.net/read.php?tid=13139
somewhere there was an analysis of a fluke reference circuit with the resistor's tempco and drift factor
included in the calculations, it was very interesting but I cannot find it anymore.

As for the oven itself for fun I have built a pair of small ovens, with discrete parts, their stability is
decent, less than half a degree, probably less, but I have only a pair of cheap lcd fridge thermometers to
check the temperature, to be sure I should use good temperature sensor and log it for some weeks...
And dont want to, the fluke 343 is more than enough for me, and probably I'll never be able to match or
surpass it

I can answer to my question:

do the 2001/2002 share the same voltage reference
with the "economic" 2000, if not what reference do they use ?

from Andreas Jahn:

Keithley 2000, 2001 and 2010 use the LM399
Keithley 2002 uses the LTZ1000A

Mitch Van Ochten notes that:

The Keithley 2002 Service Manual says it uses a "IC, ULTRA PRECISION REF,
LT21000A (PKG H8)".

Now I'm curious to see the LT21000A marking on the part

Fabio.

[Bored@Work]

Has anyone ever done this before, i.e., built a modern reference IC into a home-brew oven. Obviously, if you do, you first have to have an actual need to to that.

The worst, not the best, DIY "oven" I once did was for a crystal, to make a very bad OCXO.

The heating element, the temperature sensor and the temperature controller were made from one single component, a 7805 linear regulator.

Genuine 7805s have thermal protection. The idea was to short the output of a 7805, let it heat up, and let the thermal protection do its thing, so it would cut off the output. Then the 7805 cools down, at some point the thermal protection turned off again, output is turned on again ... So much for the theory.

In practice, I fond 7805s made in China that didn't have any thermal protection at all. Instant magic smoke. Those that had thermal protection still got very hot, no surprise, since they are specified for a junction temperature of 125 °C.  But I didn't expect them to get them as hot as they got. When you measure 150 °C on the outside, and take thermal transfer into account then the thermal shutdown must have happened beyond a junction temperature of 150 °C, uuuhm ...

Of course, heating a crystal to 150 °C is not a good idea. And using a normal cut crystal for an OCXO is also not the best thing to do. In the end it was fun, but a good OCXO it wasn't
 

[SeanB]

I did make a crystal oven years ago, using I cannot remember how, but it did work well enough, keeping the crystal at around 60C. Cardboard and styrene insulation made it pretty low power. It is still in the function generator, working well enough still.

[bingo600]

Since you got the keithley dmm. IMMEDIATELY change all the caps in the power supply. I have seen at  least 5 keithley multimeters 2001 / 2002 with shorted out electrolytics. It fries the motherboard before the fuse pops.... I salvaged the lm199's so i got a few of those laying around.

I just got one of these DMM's , same revision as Fabio's.

and was thinking about changing the caps. As per fe's warning
Does that apply to both the upper & lower boards ?

I suppose it's all the "Big caps" , i have just lifted the "lid off" and noticed the caps are nichicon ..
Arent they supposed to be ok ?

/Bingo

[kxenos]

To anyone with 20xx DMM: In my 2000 unit, the buzzer seems to have a prob: The frequency doesn't remain constant. I think it starts off a higher frequency and during the first 0.1 - 0.5 seconds it gradually falls to a steady freq. that is kept from then on. Have you noticed the same behavior?

[kxenos]

...anyone please??

[muvideo]

Hello, my buzzer tone is constant, maybe your buzzer is broken or dirty.

[bingo600]

Since you got the keithley dmm. IMMEDIATELY change all the caps in the power supply. I have seen at  least 5 keithley multimeters 2001 / 2002 with shorted out electrolytics. It fries the motherboard before the fuse pops.... I salvaged the lm199's so i got a few of those laying around.

I just got one of these DMM's , same revision as Fabio's.

and was thinking about changing the caps. As per fe's warning
Does that apply to both the upper & lower boards ?

I suppose it's all the "Big caps" , i have just lifted the "lid off" and noticed the caps are nichicon ..
Arent they supposed to be ok ?

/Bingo

Well to answer my own Q'

I was just told by an EE friend i have that : "This is my best translation ..."

Nichicon apparently in 2002 bought some Electrolyte fluid from a "Scam company".
It was used in the HM & HN series of caps. They made the bad caps until 2004.

The problem with the bad electrolyte was that it became conducting over time ,
and when there was voltage applied it "rusted/destroyed" the foil with a bad result.
Sometimes there went thermal runaway in the cap , followed by a bang.

He said they can look ok but still be bad ....

And if they get hot they're bad for sure.

So i guess FE has got a point ....

/Bingo

Uploading an archive of pics from my 2015 , just arrived from our friend KJDS 

Initial measurements with a Geller 10v Ref (2 hours warmup)

Code: [Select]

Agilent 34401A       : 09.99978  (6 digit fast)
Keithley 2015          : 09.99992  (MED)
HP 3478A                : 10.0008

It's even more spot on than my 34401A ....