3458A - black edition

[niner_007]

I have seen new A1, A3 and A5 board today. Great work 

-branadic-

Do tell is it all SMD? did they change the ADC ASIC hybrid? is the PCB multilayer?

[Andreas]

I have seen new A1, A3 and A5 board today. Great work 

-branadic-

photos please

with best regards

Andreas

[Kjelt]

Since yesterday, 1st December, only the Black edition 3458A is offered, worldwide.
The old white instrument as well as the EU stock have been removed from all of their websites.
Frank

They take that "black friday" thing way to serious at Keysight 

[SilverSolder]

The excitement is building -  Who will be first with the scoop?   

[branadic]

Sorry, no pictures that I can post.

-branadic-

[Messtechniker]

Meilhaus is advertising this thing as having "improved long term stability".
Just wondering how KS did it.

[Kleinstein]

Improved stability could be from a lower temperature of LTZ1000. The original had a rather high temperature. Such a change may effect the maximum environmental temperature a little. With a few lower power parts (e.g. FPGA, CPU part) the total power consumption could be a little lower and thus less internal heating and warming. On the original board some resistors used for current compensation at the reference are also quite close to the reference and cause heating, where it is not wanted. A new board could have improved on this too.

So while the 3458 design was a good and successful there are still a few odd and not so good design decisions.  Improvements in the thermal design could be one of the few changes they have to consider. The thermals will change with new and more SMD boards anyway.
Other points are hard to change without loosing full compatibility.

[Edwin G. Pettis]

The main reason the 3458As had 15K temperature ratios is because the vast majority of these units were in racks where as many as 5 3458As were mounted on top of each other plus other equipment and the ambient temperatures in these racks very often exceeded 50°C.  A lower temperature setting for the LTZ would have resulted in loss of temperature regulation, a big no-no in labs and test facilities.  It just wasn't worth the cost and effort to put lower value resistors in units that weren't going into such hot environments.  You guys have the option to reduce your LTZ ratios for lower temperatures.  The stability specs for the 3458A are quoted for the 15K ratio as it is well known, the LTZ circuit has very low drift with reasonably low TCR components and reducing those TCRs further does not result in significant improvement in drift specs, you have to look elsewhere for that.   Selection of the LTZ itself is the only way to get the lowest long term drift rate and that is with a long term burn-in which is 18 months at KS (depending on the target long term drift rate), those are then graded into 8PPM, 4 PPM and 2 PPM/year finished boards, there are no shortcuts.

SMD is not the best component to be used in the LTZ circuits, either best metal foil or precision wire wound should be used for top performance and low noise, SMD fits neither of those requirements.

[Henrik_V]

My information is that due to reduced power consumption of the new boards (mainly the digital parts I asume) the internal temperature if about 4K lower than in the old units.

[Edwin G. Pettis]

I agree that in theory a 4°C internal drop might help, the point I was making is that the majority of 3458As go into a rather hot rack with multiple instruments generating heat so in that particular case, you cannot drop the 15K ratio resistor to a lower value.  Now if the 3458A is sitting by itself with no other heat sources adding to the ambient temperature (say <40°C), plus the 3458A is running slightly cooler due to less self heating, the other circuitry doing the measurements may exhibit a little lower drift and contribute to a slightly better drift component.

Any of your 3458As sitting in the open on a table or shelf could possibly show a slightly lower drift overall compared to an earlier 3458A but the ambient temperature is still going to control the overall drift as usual.  It has been proven that a fair number of 3458As that have the higher yearly drift specification will in time achieve the same 2 PPM/year drift as the selected models, it just takes longer for the LTZ module to settle in and some never do get that stable.  The LM399A can also achieve very low drift rates with time, unfortunately time does not improve noise to any significant degree.

[jklasdf]

Just got an email from Keysight about the new 3458A today. It does seem like the main advertised features are just RoHS compliance, the new industrial design, and no *worse* performance than the original. Seems pretty conservative. My guess is they probably would have tried to change as little as possible with the reference and ADC other than for parts obsolescence, even if it could theoretically give better performance.

[Dr. Frank]

KS needed to keep all versions since 1989 compatible, and especially the specification must not change.
I learnt that the reason for that is, that the 3458A is mainly intended for industrial use, i.e. inside manufacturing and testing lines at elevated temperatures, and these production lines are validated / homologized /audited. If you would change the 3458A specification, all these manufacturing lines would have to be re-audited, if an instrument would be replaced.
HP did not even improve / adapt the specifications, when several very obvious changes were made in the past, like the much better 40kOhm reference resistor in about 1995.

So although the new Black Edition really has many improvements and redesigns, formally / legally it's fully backward compatible.
For the very same reason, firmware is unchanged, as well as the interface.

On our Metrology Meeting in June, the KS representative already told us, that the DCV noise was halved, and that the DCI ranges are about 10 times more stable, and that the instrument is more stable overall.
Maybe the engineering group in charge for this redesign used  better shunt resistors, probably similar to the 8085A, from Vishay.

Then, all these terminated components, like the EL2008 comparator had to be replaced by other, improved circuits and components. All that cannot be done by simply replacing through hole components by SMD, you have to completely redesign everything.
If you think about that further, 8051 and 68000 processors are also obsolete, and a compatible, modern solution had to be chosen.. not difficult to imagine how this was accomplished, and why now the 128kB memory is included.

Frank

[TiN]

Let's make a fundraiser, and buy one for science teardown and deep testing.
If we get 20-40 really interested people, it's probably could be enough to buy new unit. I'd chip in as well.
Then setup Internet to GPIB interface and let everyone online to play with instrument remotely  like I like to do in friends lab. 
Yes, I know we wouldn't get enough people to spend money just for making $11k teardown.

[SilverSolder]

Perhaps Keysight would provide a loaner unit for review, if asked nicely?

[TiN]

I did try to ask nicely, but to no avail. Not even "we not interested, thank you for your inquiry". But people did say that "I am not nice" to me before, so ymmv. 
Also loaners often would mean no teardown, no extended tests (e.g. drift over a year).

[SilverSolder]

Keysight might give Dave first dibs on a review, in recognition of him being behind the collective awesomeness that is the eevblog and its forum.

Down the road, something more may be possible.  It is is obviously good advertising for Keysight and I doubt their marketing department was born yesterday!

[MiDi]

Let's make a fundraiser, and buy one for science teardown and deep testing.
If we get 20-40 really interested people, it's probably could be enough to buy new unit. I'd chip in as well.
Then setup Internet to GPIB interface and let everyone online to play with instrument remotely  like I like to do in friends lab. 
Yes, I know we wouldn't get enough people to spend money just for making $11k teardown.

Interesting idea.
If it was announced that the unit is drawn to one fundraiser, this should get more attention.
Before it goes to winner a teardown and 1 year test is done by e.g Dave, TiN.
It would be fair, if the chance to win depends on amount of money spent.
 

[Kjelt]

KS needed to keep all versions since 1989 compatible, and especially the specification must not change.
I learnt that the reason for that is, that the 3458A is mainly intended for industrial use, i.e. inside manufacturing and testing lines at elevated temperatures, and these production lines are validated / homologized /audited. If you would change the 3458A specification, all these manufacturing lines would have to be re-audited, if an instrument would be replaced.

Thanks very informative.
What I do not understand is the marketing strategy in this case.
They can label it twice and sell at different prices, one as a 3458A with up price for the industrial cases you mentioned, one as 3458B with the improved specifications listed for new customers.

Competition is not standing still either, if they now have better specs they should list them to stay in front.

[Kleinstein]

The big question is how they have replaced those special HP asics, with switches and resistor networks inside.

Any reason they simply can't make another run of the original design?  Advances in manufacturing might make this affordable over a redesign.

It shouldn't be a problem as the design is well proven and should be 'known' but I don't know if it's possible to use a modern fab to make an *old* design, would the processes be compatible (genuinely don't know, I'd be interested to hear opinions)?

A completely new design takes quite some effort and would likely not be fully compatible. At the sub ppm level odd things (e.g. leakage, board DA, acoustic coupling) can become important and it needs a lot of testing to verify the performance. The 30 years of proven use is a big asset for the old 3458.

It could still make sense to have a completely new design, especially for the industrial use - a more targeted product (e.g. less priority on the AC part) could be a lot cheaper to produce and lower power.  A new design would likely be more like the DMM7510 with a separate slow and fast ADC. This simplifies things a lot. Despite the effort the 3458 ADC in the fast mode is still no match for modern SAR chips. A multi-slope ADC for only the slower part (e.g. 1 ms and longer integration) is easier, but would still need some experience and knowledge beyond the text books. However a design with 2 ADCs would not be fully compatible anymore: it can be better in many aspects but could be slightly lower specs in the intermediate range.  The ADC in Keysights 34410, 34470 and similar is quite different from the 3458: good and maybe even better at higher speed, but limited (still sufficient for there use) linearity.

P.S.
The original 3458 ADC kind of relies on high performance JFET switching and thus the special ASIC. It still possible to produce such a thing, but it still is expensive and there may not be much improvement from the last 30 years. A different design could use cheaper MOS switches possibly even of the shelf parts.

[bsdphk]

Most likely not.  HP's "silicon on saphire" process was quite special.

[MiDi]

For those who did not follow the other Thread (of the many that are there): Keysight gave some details

[syau]

Looking forward for Daniel to send an unit to TiN 

[notfaded1]

Idk about it not having to be somewhat redesigned... Some things just aren't practical anymore... If it is someone is going to be able to beat it I bet... and sooner than later.  I'm sure fluke is working on it.  So much can be done with new generation ASICs and combined with general purpose processors the stuff that can be done in software alone is amazing.  It's what's made all the new Cisco network switches able to do software defined networking.  This year the big news was 400G Ethernet for God's sake.  I work with custom ASIC hardware engineers every day and the technology is so far beyond 5 years ago it's incredible.  I have a hard time believing time can just keep standing still...

Bill

[TiN]

notfaded1
While ASIC advancement for digital stuff is indeed progress with light speeds, that does not apply to analog domain. And no matter of digital power, math or smart logic, none of that can help much to make better analog performance front-end/ADC device, when you already pushing the limits at ambient temperatures. Digital algorithms are as good, as your input data is. If you feed garbage at the input, you only get nice looking garbage at the output . Even new Fluke meters with fancy touchscreens and pretty UI still will have tough time matching even 3458A original design from 1989. Call me a fanboi, but I'm not going to replace my old 3458A's any time soon. It's hard to imagine, but digital ASICs are VERY different to high-performance analog ASICs which often need special materials/processes and manufactured by 180 nm or larger nodes. This tech is on the opposite very difficult and expensive spectrum of the chip design, compared to fast and dense digital stuff.

Another limiter factor is voltage reference. No matter old meter, or new one, they already are reaching limits of what noise/long-term stability can be achieved using very best reference money can buy, such as LTZ1000A. And even that needed special selection and aging, before device can be put into each specific DMM.

Adding fancy Ethernet, high power hungry TFT with touchscreen and cloud-data uploading and blockchain to the meter like 3458A isn't very important for people who spend $11k+ on this instrument. Many applications and customers might still use 3458A as very fast and accurate DMM in automated test jigs. For all they care, there could be no front panel version and it would still sell just fine.

Only wish I would improve on 3458A probably would be better ohm functionality and electrically-switched front/rear input, so it could be used as direct comparator in metrology labs. But it's not big deal, one can always use external scanner and 5720A calibrator for advanced needs.

[FrankBuss]

The neat thing about the Pi, is that you can set it up ready to go and just distribute the image (sdcard) or the Pi itself, guaranteed to Plug and Play, and just work.

Someone should make a Ubuntu based distro for T&M

There is already a project for the Raspberry Pi for many instruments:

https://www.eevblog.com/forum/metrology/raspberry-pi23-logging-platform-for-voltnuts/

It has some Python examples, for the 3458A as well, see here:

https://github.com/PlesaEEVBlog/RPi_LogNut/tree/master/python

I used this with my Fluke 8842A and worked nicely. But meanwhile I'm using Debian Linux on my main desktop system for daily work, and managed to install the GPIB driver on it with the help of the installation scripts on the voltnuts repository. Now I can just run the Python scripts from my desktop, which is even easier than the additional step to the Raspberry Pi.

Example how I recently used it, with an updated version of my voltage logging script, and then using Google Sheet to create a diagram of the recorded values:

https://www.eevblog.com/forum/dodgy-technology/9900-mah-18650-accus/msg2725094/#msg2725094