Comparing Agilent InfiniiVision 2000 and 3000 X-Series Oscilloscopes

[Faith]

They went with 4Mpts because that's what size memory is built onto the MZ4 die, that's what makes it crazy fast.
If the higher power scope use more and the same ASIC then they must be bypassing the internal memory and using external.

Yep, and looking at the specs of the higher end Keysight scopes with more memory it seems that their waveform update rate is quite a far cry away from the 1,000,000 wfms/s we see on those with only 4M memory points.

Their sampling rates are still pretty insane though.

[Wuerstchenhund]

BTW, how old is the 2000/3000 series now? 
Remember when it used to be untouchable?!

What do you mean with "untouchable"?

[Faith]

What do you mean with "untouchable"?

I would say that the 3000T has always been quite touchable

Still though, as mentioned earlier, newer models don't necessarily make older models useless. But the market has definitely become really competitive and thus all the more interesting.

Let's see where the wind takes me though!~

[Someone]

We may have to wait a while for the MZ5. Clearly the MZ4 can be used in pairs: I wonder if an architecture using four of them is possible? Also presumably, if the ASICs going into current production 2000s are fully functional, a firmware offering 2Mts might be possible.

I doubt there are as many imperfect MZ4 ASICs as you imagine, after all as others already mentioned the 65nm process is ancient and already has been when the DSOX came out back in 2011. It's a well controlled process with a very high yield rate.

65nm was entering mainstream in 2011:
https://eda360insider.wordpress.com/2011/09/07/where-is-the-mainstream-ic-process-technology-today-28nm-40nm-65nm/
But the megazoom IV ASIC is reportedly on the 90nm node:
http://www.embedded.com/electronics-news/4213152/Agilent-uses-new-ASIC-in-MSO-market-attack
Which was mainstream for fully digital designs of the era. Just because the bleeding edge is 1-2 nodes ahead doesnt make a process ancient.

[Faith]

Just because the bleeding edge is 1-2 nodes ahead doesnt make a process ancient.

I think the original concern was why the 2000A was more severely limited in memory than the 3000A. And while binning could have been a reason, marketing is more likely as yields should've been really good by then and especially now.

[nfmax]

The big advantage of putting the memory on the same die as the logic is you can increase the memory 'bus width' between the two to crazy levels - many thousands of bits if you want - even for relatively small memory sizes. The means you don't need such a fast memory process, and it massively increases the memory bandwidth, which is generally the limiting factor in DSO applications. On the other hand, if you have an architecture with lots of fast RAM chips and FPGAs, like the typical higher end scope, then you get the increased memory bandwidth 'for free' as there are so many RAM chips each with their own data bus. Of course this costs more, and is much more power hungry, but these are acceptable in a high-end instrument. So the all-in-one ASIC design is better suited to lower tier designs (which are likely to sell in much larger volumes anyway)

Entry-level instruments using commodity DDR3 memory are limited by its 64-bit bus width. So you can easily & cheaply provide a lot of memory, but the waveform update rate is likely to be less than stellar, and if you aren't careful the UI will be sluggish as a result. or you can just offer a few thousand points of memory, sacrificing memory depth for waveform update rate & responsiveness. Current fashion in entry level scopes seems to be to cram as much memory in as possible for spec sheet bragging rights. I have no personal experience of using one of these scopes, but I expect if you wind the memory depth up to '11' the UI will get a bit treacly.

Comparing the two series, the 3000's offer a maximum of 4Mpts memory, while the 2000's offer only 1Mpt. But it seems there are actually two MZ4 ASICs in the 3000 and one in 2000, so you might expect a 2:1 ratio in maximum memory. That got me wondering if the 2000 was originally designed to be able to use part-good die, as well as lower speed bins (it has 2Gs/s compared to the 3000A's 4Gs/s)?

Nope, it's the exact same good die. The limited memory on the 2000 is purely a marketing spec decision. They always reserved the right to up the memory later via a software update to compete with competition.

Dave - you are right, IIRC the initial software only supported 100kPts, and no segmented memory either (and when segmented memory came out, it was originally limited to 25 segments). Who knows what goodies may still be in store for us, then?

[Wuerstchenhund]

They went with 4Mpts because that's what size memory is built onto the MZ4 die, that's what makes it crazy fast.
If the higher power scope use more and the same ASIC then they must be bypassing the internal memory and using external.

Yep, and looking at the specs of the higher end Keysight scopes with more memory it seems that their waveform update rate is quite a far cry away from the 1,000,000 wfms/s we see on those with only 4M memory points.

Yes, with Keysight Unfortunately while they do have the fastest entry-level scopes they don't when it comes to further up the ladder.

Their sampling rates are still pretty insane though.

Larger bandwidth requires higher sample rates. The current sample rate maximum are 240GSa/s but again, not from keysight

[Wuerstchenhund]

65nm was entering mainstream in 2011:
https://eda360insider.wordpress.com/2011/09/07/where-is-the-mainstream-ic-process-technology-today-28nm-40nm-65nm/

The article, which is from 2001 and mostly concerned with the 28nm process, says nothing of that kind. All it says that Gartner found that in 2011 65nm was considered mainstream. In fact, if you look at the graph, it shows that 65nm had a similar share in 2010 and 2009 (unfortunately the graph doesn't go further back than 2009), which makes the idea that it became "mainstream" in 2011 even more absurd.

And all this doesn't change the fact that 65nm is still a pretty old process:
https://en.wikipedia.org/wiki/65_nanometer

Besides, there are many designs still made in 90nm and 120nm but that doesn't make these processes less old. Even more so, quite often the fact that they are old is the reason a designer goes for them, as that makes them cheap (in regard to the process costs, not necessarily the material cost)  and well controlled.

[Wuerstchenhund]

The big advantage of putting the memory on the same die as the logic is you can increase the memory 'bus width' between the two to crazy levels - many thousands of bits if you want - even for relatively small memory sizes. The means you don't need such a fast memory process, and it massively increases the memory bandwidth, which is generally the limiting factor in DSO applications.

Yes, you could. Not sure if that is true for MZ4, though.

On the other hand, if you have an architecture with lots of fast RAM chips and FPGAs, like the typical higher end scope, then you get the increased memory bandwidth 'for free' as there are so many RAM chips each with their own data bus. Of course this costs more, and is much more power hungry, but these are acceptable in a high-end instrument. So the all-in-one ASIC design is better suited to lower tier designs (which are likely to sell in much larger volumes anyway)

That depends on the design. Most high-end scopes are pretty slow when it comes to update rates, which don't really matter that much in that class as they do in the low-end segment. Especially designs that do their processing in ASICs are generally relatively slow in processing the large amounts of data a modern high-end scope produces (other manufacturers use standard x86/x64 intel processors to get better performance).

Entry-level instruments using commodity DDR3 memory are limited by its 64-bit bus width. So you can easily & cheaply provide a lot of memory, but the waveform update rate is likely to be less than stellar, and if you aren't careful the UI will be sluggish as a result.

Yes, but at least it is cheap. Which is the prime objective for pretty much any B-brand.

or you can just offer a few thousand points of memory, sacrificing memory depth for waveform update rate & responsiveness.

As explained further up, there is a fixed limit to how often a scope can update when using large memory. Even MegaZoom can't change that, it pretty much just cheats itself out of it by using a small sample size during operation.

Responsiveness is an UI issue. Tek is a good example how not to do it as on its scopes the UI locks up during longer operations. Other manufacturers maintain a responsive UI at all operations.

Current fashion in entry level scopes seems to be to cram as much memory in as possible for spec sheet bragging rights. I have no personal experience of using one of these scopes, but I expect if you wind the memory depth up to '11' the UI will get a bit treacly.

Yes but again that is because especially the B-brand scopes are targeted at the lowest possible price point. They also can't make a lot of use of their large memories, i.e. due to the lack of proper analysis tools. And despite the large memory, FFT is often poor as well (i.e. some 4kpts on Rigol DS2000 Series scopes if I remember right).

[nfmax]

There is probably a distinction to be made between the 'B brand' entry level scopes like the DS 1054Z and the 'A brand' entry level scopes like the Keysight 2000 & 3000 scopes. The former are targeted at hobbyists (& Shenzhen backstreet workshops) while the latter are aimed at general-purpose lab hack scopes - the sort of scope that spends weeks hooked up to a prototype repeating the same measurement, or even as part of manufacturing test. Overlapping both are the educational scopes, where Tektonix still rules because of their name and 'flexible' pricing. The Keysight 1000 range is aimed there too, but without much traction.

As you move up the ladder then scopes become much more specialised: a 100GHz LeCroy would be a poor choice when troubleshooting a DDR4 interface, while a Keysight S series can't achieve the bandwidth needed for fast opto-electronics. Though you might be able to use either, it would be a compromise.

[Wuerstchenhund]

There is probably a distinction to be made between the 'B brand' entry level scopes like the DS 1054Z and the 'A brand' entry level scopes like the Keysight 2000 & 3000 scopes. The former are targeted at hobbyists (& Shenzhen backstreet workshops) while the latter are aimed at general-purpose lab hack scopes - the sort of scope that spends weeks hooked up to a prototype repeating the same measurement, or even as part of manufacturing test.

It's not just entry-level B-brand scopes, though. Just look at Rigols DS6000, a $9k scope that by its price would be solidly placed in the mid-range. However, it is neither in performance, feature or maturity (apparently basic things like ETS sampling are still broken).

Overlapping both are the educational scopes, where Tektonix still rules because of their name and 'flexible' pricing.

And to some extend to the fact that many educators don't get out very often and from their youth pretty much only know Tek (analog) scopes.

The Keysight 1000 range is aimed there too, but without much traction.

Well, the DSOX2k isn't a complete stranger to the edu market, I believe Keysight has even some bundles (with special pricing) for them.

As you move up the ladder then scopes become much more specialised: a 100GHz LeCroy would be a poor choice when troubleshooting a DDR4 interface, while a Keysight S series can't achieve the bandwidth needed for fast opto-electronics. Though you might be able to use either, it would be a compromise.

I agree with your example, but that is pretty much only because that 100Ghz LabMaster is a very special and unique scope. In general, with a modern high-end scope you can do pretty much most things you can do with your entry-level scope. As mentioned in another thread my daily work horses are highend scopes, and I use them even for basic measurement because going to fetch some smaller scope is combersome and modern high-end scopes do simple jobs more than good enough.

[nfmax]

I agree with your example, but that is pretty much only because that 100Ghz LabMaster is a very special and unique scope. In general, with a modern high-end scope you can do pretty much most things you can do with your entry-level scope. As mentioned in another thread my daily work horses are highend scopes, and I use them even for basic measurement because going to fetch some smaller scope is combersome and modern high-end scopes do simple jobs more than good enough.

You are lucky! In many situations, the high-end scope is a shared resource, so a hack scope that doesn't have 'legs' has a lot of advantages!

[EEVblog]

I think the original concern was why the 2000A was more severely limited in memory than the 3000A. And while binning could have been a reason, marketing is more likely as yields should've been really good by then and especially now.

I can guarantee you it's entirely marketing driven, I've heard this direct from head people at Agilent at the time. It is no secret.
And IIRC they increases the memory in the 2000 series as per their plan last year or so I think it was?
The waveform update rate is also limited by marketing decision, the 2000x is capable of the same update rate as the 3000x. Same thing with serial decoding, they even left the unused button on the front panel "just in case".

[EEVblog]

Overlapping both are the educational scopes, where Tektonix still rules because of their name and 'flexible' pricing.

And to some extend to the fact that many educators don't get out very often and from their youth pretty much only know Tek (analog) scopes.

Educational scopes are driven by the educational course materials that come with them.
Look at the news Tek TBS2000 series just released, it is chock full of education stuff built in that tie into the course lessons available, directly on the scope display.

The Keysight 1000 range is aimed there too, but without much traction.

Well, the DSOX2k isn't a complete stranger to the edu market, I believe Keysight has even some bundles (with special pricing) for them.

Keysight offer educational course material and educational modes on the scope much like Tek does.

[EEVblog]

Responsiveness is an UI issue. Tek is a good example how not to do it as on its scopes the UI locks up during longer operations. Other manufacturers maintain a responsive UI at all operations.

This is where the MZ4 ASIC shines.
It's not just acquisition but it does the display processing too, along with FFT and serial decoding in hardware, everything on the one chip. That's why the Keysight X-series never slows down when you enable serial decode, FFT, or use the controls or anything else.

This is also a downside as seen on the 4000X series when they used the larger screen, the resolution was crippled because of the fixed display resolution inside the MZ4 ASIC. They could not increase the display window size without ditching the entire MZ4 ASIC. Any newer Keysight scopes will have to use a new MZ ASIC to overcome the display limitation and the memory size limitation. Although you can bet your bottom dollar they will continue to milk the MZ4 ASIC because the investment has been made and recouped there, they have the potential to churn out a high-ish performance hardware cheaply.

MZ1 was 1996
MZ4 was 2011
So 3 new ASIC's in 15 years. We are almost overdue for another one one.

[nfmax]

This is also a downside as seen on the 4000X series when they used the larger screen, the resolution was crippled because of the fixed display resolution inside the MZ4 ASIC. They could not increase the display window size without ditching the entire MZ4 ASIC. Any newer Keysight scopes will have to use a new MZ ASIC to overcome the display limitation and the memory size limitation. Although you can bet your bottom dollar they will continue to milk the MZ4 ASIC because the investment has been made and recouped there, they have the potential to churn out a high-ish performance hardware cheaply.

MZ1 was 1996
MZ4 was 2011
So 3 new ASIC's in 15 years. We are almost overdue for another one one.

Because of the ever-growing acquisition memory size, and the increasing capability of FPGAs, the ASIC sweet spot has been moving toward entry-level scopes over the period. It will be interesting to see where MZ5 first appears! 

[Someone]

65nm was entering mainstream in 2011:
https://eda360insider.wordpress.com/2011/09/07/where-is-the-mainstream-ic-process-technology-today-28nm-40nm-65nm/

The article, which is from 2001 and mostly concerned with the 28nm process, says nothing of that kind. All it says that Gartner found that in 2011 65nm was considered mainstream. In fact, if you look at the graph, it shows that 65nm had a similar share in 2010 and 2009 (unfortunately the graph doesn't go further back than 2009), which makes the idea that it became "mainstream" in 2011 even more absurd.

And all this doesn't change the fact that 65nm is still a pretty old process:
https://en.wikipedia.org/wiki/65_nanometer

Besides, there are many designs still made in 90nm and 120nm but that doesn't make these processes less old. Even more so, quite often the fact that they are old is the reason a designer goes for them, as that makes them cheap (in regard to the process costs, not necessarily the material cost)  and well controlled.

The article is from 2011, the year in question and shows a single graph with an estimated breakdown of the node shares. For 2011 just 25% of chip fabrication was on smaller nodes than 90nm, that puts 90nm right in mainstream despite its share declining. Very few low-mid volume ASIC customers use new nodes until the libraries are mature and proven which makes 90nm a typical choice for an ASIC publicly released/presented in 2011, considering the long development time involved for these test and measurement products.

[nfmax]

It's not just entry-level B-brand scopes, though. Just look at Rigols DS6000, a $9k scope that by its price would be solidly placed in the mid-range. However, it is neither in performance, feature or maturity (apparently basic things like ETS sampling are still broken).

Well that's just an uncompetitive scope that no one is likely to be considering, unless forced to. An exit-level design, maybe?

[Keysight DanielBogdanoff]

I doubt it's the exactly same ASIC, but yes I'd guess the memory is external.

Not the same ASIC. We actually use a combo of in-house ASICs for the acquisition systems on various scopes. The "MegaZoom" sticker basically means it's using an in-house processing ASIC.

I can guarantee you it's entirely marketing driven, I've heard this direct from head people at Agilent at the time. It is no secret.

Yup. We knew going in that the MZ4 would be overkill for the 2000X, but it made more sense to use one ASIC across the family instead of designing multiple.

Comparing the two series, the 3000's offer a maximum of 4Mpts memory, while the 2000's offer only 1Mpt. But it seems there are actually two MZ4 ASICs in the 3000 and one in 2000, so you might expect a 2:1 ratio in maximum memory. That got me wondering if the 2000 was originally designed to be able to use part-good die, as well as lower speed bins (it has 2Gs/s compared to the 3000A's 4Gs/s)?

Having two MZ4 is not about memory depth, it's more that we're using certain functions from each MZ4. For example, on the 2000 you can't do protocol decoding and use digital channels at the same time. Why? A single ASIC can only do one of those two things at a time. Having two MZ4s in the 3000+ gives us the ability to do digital channels on one ASIC, decoding on the other.

Also, AFAIK our yields are pretty much standard for their given geometries and chip sizes.

[TheSteve]

I can guarantee you it's entirely marketing driven, I've heard this direct from head people at Agilent at the time. It is no secret.
And IIRC they increases the memory in the 2000 series as per their plan last year or so I think it was?
The waveform update rate is also limited by marketing decision, the 2000x is capable of the same update rate as the 3000x. Same thing with serial decoding, they even left the unused button on the front panel "just in case".

I believe this to be true but keep in mind these are the same people who told you the 350/500 MHz model used a much more expensive front end then the 100/200 MHz model.

I always wondered why my DSOX3K didn't give any indication of the sample memory on screen - guess they figure you don't need to know, and maybe they don't want you to know.
Either way I think it is an amazing scope and am very happy with it.

[Wuerstchenhund]

It's not just entry-level B-brand scopes, though. Just look at Rigols DS6000, a $9k scope that by its price would be solidly placed in the mid-range. However, it is neither in performance, feature or maturity (apparently basic things like ETS sampling are still broken).

Well that's just an uncompetitive scope that no one is likely to be considering, unless forced to.

You say this but there was a forum member who bought one after asking here and getting mostly negative feedback regarding that scope, only to return it later because it was so poor

I believe in the animal world this is called "learning through pain"