Tearing Into A $1.3 Million Oscilloscope

Most hackers are rankled by those “Warranty Void If Broken” seals on the sides of new test equipment. Even if they’re illegal, they at least put the thought in your head that the space inside your new gear is off-limits, and that prevents you from taking a look at what’s inside. Simply unacceptable.

[Shahriar] has no fear of such labels and tears into just about everything that comes across his bench. Including, most recently, a $1.3 million 110-GHz oscilloscope from Keysight. It’s a teardown that few of us will ever get the chance to do, and fewer still would be brave enough to attempt. Thankfully he does, and the teardown video below shows off the remarkable engineering that went into this monster.

The numbers boggle the mind. Apart from the raw bandwidth, this is a four-channel scope (althought the unit [Shahriar] tested is a two-channel) that doesn’t split its bandwidth across channels. The sampling rate is 256 GS/s and the architecture is 10-bits, so this thing is dealing with 10 terabits per second. We found the extra thick PCBs, which are perhaps 32-layer boards, to be especially interesting, and [Shariar]’s tour of the front end was fascinating.

It all sounds like black magic at first, but he really makes the technology approachable, and his appreciation for fine engineering is obvious. If you’ve got even a passing interest in RF electronics you should check it out. You might want to brush up on microwave topics first, though; this Doppler radar teardown might help.

61 thoughts on “Tearing Into A $1.3 Million Oscilloscope

  1. impressive, but as far as i understand, he did not actually take it apart, he got some extra boards from the manufacturer for the review, so he does not need to open and take the boards from the working unit.

        1. Some of us can’t hear/process speech as easily as text, add in video, instead of still images, and at best it’s a case of trying to rewind/pause/fast forward to try and figure out what is being said about what.

          The end result, a frustrating waste of time.

        2. I’d rather have it documented like it’s a report that’s ready to be printed out or shown like a old fashioned PowerPoint.

          It’s simply easier to digest and more to the point with minimal fluff and fuzz.

          Videos should be restricted to something that can only be demonstrated “in action” and that’s it.

        3. I’m the same. I just don’t watch videos. They’re not a good way to learn information. They might be useful to learn a skill, if they’re demonstrating some sort of manual technique or something. And video is a good medium for entertainment, of course. But for information, words and appropriate diagrams and pictures are how grown-ups learn. They’re also hundreds of times smaller.

          When HAD post links to videos it gets right on my tits. There should be some sort of flag or something so I can skip them.

        4. I’m with you john. Something about “gift-horse” and “mouth” jumps to mind here…

          Thanks for the free edu-tainment Shahriar. I have no need to interface with such gear, but it is really interesting to see your exploration of the sample path.

  2. Not much of a teardown (there’s definitely limits what a manufacturer will let you do with one of their still unfinished prototypes I suppose), but still a good video. It’s interesting to see some of what goes into equipment with this level of performance.

  3. I can think of something better to spend 1.3 million on. It would not be this overpriced tool. Sorry I know it’s a shit hot scope but it’s a joke of a price. But I sure a big company must make good use of it to make there money back and more. Not for us with little workshops.

    1. Small scale narrow thinking.
      If you are designing a producht that needs a 110GHz scope and have the budget to spen 1.3M on an oscilloscope you are likely in a market in which billions can be made.

    2. The price is in the R&D, several ASICs had to be developed to make this happen. Naturally the top-of-the-line model will cost as much as this to justify the whole development. For the institutions who need it, it makes sense.

      1. “Naturally the top-of-the-line model will cost as much as this to justify the whole development.”

        Which means also that over time, the now functional ASICs will trickle down into the less expensive units, giving a burst of extra bandwidth and features.

  4. If you need it, you need it. If you were developing RF gear in that frequency you would need it. This is not home gamer stuff. A lot of the high end test gear is leased for specific projects anyway.

    1. If you were on my engineering team developing RF gear at that frequency and told me you need that scope to do your job, I’d probably fire you.

      Now, if we were developing 100 Gbps transceivers, then maybe you’ve got something.

      1. >Now, if we were developing 100 Gbps transceivers, then maybe you’ve got something.

        Except you’ll *need* quite a bit more than 100GHz bandwidth scope to look at 100Gbps signals. (I know that bit rate is not frequency.) Depending on what you are looking at, it could be anywhere between 3X to 5X of the fundamental frequency.

        BTW that’s really basic stuff if you work with scopes.

        1. True, that. Sampling 101. You do generally need several times your baseband frequency content to adequately sample a signal.

          Fortunately, that didn’t stop development of 10 Gbps fiber and Ethernet 20 years ago, when even 10 GHz scopes were unobtanium.

          1. I have a HP83480A digital sampling oscilloscope mainframe which can accept two dual 50GHz electrical input modules (for a total of 4 channels at 50GHz) sitting in my basement still waiting for some modules that wouldn’t cost an arm and a leg.

            Its manual was printed in 1997 which was over 20 years ago, so I wouldn’t say they were exactly unobtainium. They still *are* expensive as hell, but when you’re developing stuff that really needs that bandwidth, you can get them.

          2. The 83480A was really “just” a 2.5 GHz scope. Its higher speed spec comes from the sampling front ends that sequentially grab multiple (nominally identical) stable waveforms with different phase delays to build up the higher “equivalent time” effective sampling rate. Still amazing for its time.

  5. Having watched the video, it sure is a nice presentation.
    Not to mention an impressive scope.

    Though, I at times wonder, if one does such high bandwidth measurements, then couldn’t one just use a real time spectrum analyzer and a multimeter to see the DC offsets? (No offense here. But the bellow idea is already how this scope works, but with a single down conversion stage that covers only a portion of the RF spectrum, instead of multiple parallel ones to cover the whole bandwidth of our scope.)

    I am curious to how far one could go, lets say one has a real time spectrum analysis bandwidth in the area of couple of GHz range, that we can offset throughout the whole RF spectrum. Lets say just for comparisons sake that we can view the same area of bandwidths as this scope mentioned above. So DC-110 GHz. (Though, the DC part might be measured with a traditional Oscilloscope directly, so our down converter doesn’t need to work that far down, so a few GHz might be good enough for our low end.)

    Then with that oscilloscope with its internal down converter + some device for measuring DC offset (a multimeter with a low pass filter. (likely built into the scope)) then how would that compare to this scope for real world measurements? (Yes, setup time would likely be longer since we need to tune our down converter into the right area so that we can even measure our signal to start with.)

    Reason for me asking this question is for the following, how often do one work in the tens of GHz range and keep content throughout the RF spectrum all the way down into the DC range?

    Since having only one down converter and a “single” ADC for sampling our data would be far cheaper then the solution to sample the whole spectrum that the scope is able to cover. Though, yes, we could need more bandwidth, so we might have two down converters, that is still cheaper then the array of down converters in the scope above. And with two, we could also offset them to different areas of interest, so that would still allow us to look at wildly different areas of our spectrum, while not needing the expense of being able to cover the full spectrum.

    1. That’s not the same thing. Sure you could use a spectrum analyzer to determine the frequency response of a circuit or device, but an oscilloscope is showing you the actual waveform of the signal from that device. Time domain vs Frequency domain…

      1. Yes, a traditional spectrum analyzer isn’t the same thing as an oscilloscope.
        And yes, time and frequency domains are different.

        But now, a real time spectrum analyzer has more in common with an oscilloscope then with a spectrum analyzer. And is actually how Keysight even manages to reach 110 GHz in the first place.

        What we are doing is sampling a bandwidth, this could be 5 MHz, or 10 GHz, really doesn’t matter. If we start at DC, then it works like a traditional oscilloscope, if it is AC coupled with a capacitor, then people still call it an oscilloscope. (Even though it suddenly has more in common with a real time spectrum analyzer. Just lacks a few parts, namely a down converter.)

        We aren’t actually looking at the frequency content of the signal, but rather the waveform of the signal after some heavy band pass filtering. Down conversion does make our lives a bit harder, but it isn’t something we can’t actually deal with.

        And I should repeat again, this is how Keysight actually have implemented their 110 GHz scope. (LeCroy used the same technique for their 100 GHz oscilloscope.)

        My question is simple, why sample everything on the spectrum, is that really needed in most applications?

        Think of a scope that samples a 10 GHz bandwidth between 90-100 GHz as a 100GHz oscilloscope that has a great amount of high pass filtering in front of it, but in that case, we could just as well use a down converter and a 10 GHz capable ADC, and that is far less resource intensive then sampling the full 100GHz. (And yes, I am greatly oversimplifying here.)

    2. The beauty of a digital down converter is that you have no IQ impairments. It’s perfect. If your measurement can withstand the impairments , then great, analog down conversion is wonderful. And real time signal analyzers are great too if you must have that real-time trigger. Other than the trigger and processing power to pull it off, a real-time analyzer is identical to the non-real-time analyzer.

  6. Not meaning to dis Shahriar, and it is a topic I find fascinating, but I have to agree video is an awful way to present this stuff, especially a whole frickin’ hour of it. Not skimmable, not searchable. Not really useful random access. Though the voiceover may be technically adequate, it’s painful waiting for the information to trickle in. You’re forced to swallow everything presented at the rate it is presented.

    I mean, 14 minutes on a single image? Even if it does have squiggles added from time to time, it’s easy to find something else to pay attention to.

    I really, really wish it were more accessible – I’d love to learn more about that scope. Sorry, I had to give up after 25 minutes.

    Sure, it’s much more work to write it down with high-quality diagrams and annotated images, but the resulting document would be far more accessible and useful, but I can see why producers like Shahriar make the video: it’s much easier to get a lot more content out cheaply. If he has an audience that accepts that, I can’t fault him for it.

  7. That’s a pretty impressive instrument: its 64 GHz ADCs use a sample period of 15.6 picoseconds. OK, that’s impressive by itself: One sample every 5 millimeters at lightspeed.

    Arranging a few of those converters to get the right phase alignment, that’s impressive too.

    But what I find literally unbelievable is the 10 bits precision it claims at that rate. To achieve that requires a clock jitter of less than 2.5 femtoseconds: the time it takes red light to travel one wavelength. This is a hundred times better than the usual “ultra low jitter” stuff us plebs can buy. *This* is the magic part.

  8. Thanks Shahriar for the excellent video. Sadly I don’t have time to watch the whole thing, I too had to go after 10 minutes or so. But don’t let that stop you.

    While I agree it isn’t my preferred medium, in fact not even close, especially when consuming technical information, it is still an excellent “teardown”. What little I could afford to watch (time, bandwidth, grey hair) was very interesting.

    summary: Thank you. Keep up the good work. I prefer shorter videos or words. I am likely not the intended audience. I am but one person.

    (before you ask, I didn’t watch all of the video)

  9. I’m GLAD it was mentioned…! Now, back to my PC, (1 or 0, pocker chip.) And maybe a more useable (for me,) portable scope to 200MHz or even less. That scope is like a yacht within a yacht… past sci phi… LOVE it, but it needs a support cast. Ulhura (sp) and that bald chick maybe. Slew rates at 27ns was as far as I got, aside from some light stuff. Let us know what you guys build with it. Outta be outta site! Maybe (eww) shrinking injectable mini-subs. Never know. But, seriously, a holographic displsy should be the next add on.

  10. That oscilloscope is a masterpiece of technology, engineering and applied R&D.
    The YT video has perfect sound, crisp image, close-ups, it even has shoots under a microscope.
    The explanations are coming from one of the finest RF/Microwave specialists (Shahriar is a PhD, EE, Researcher, University Teacher and EE Video Blogger).
    The HaD article is short and clear, mentions key features and top performance numbers, then sends straight to the source material for those who want more info.

    Simply said, everything is of an outstanding quality, with lots of knowledge given for free.
    Much appreciated, thank you!

    No comments about some comments.
    :o)

  11. I’m lucky, I am retired and can spend an hour watching this video. Congratulations to Keysight for undertaking this project, it appears to have been a massive undertaking and $1.3M seems cheap for such a capable and innovative instrument.

  12. I feel like a newbie…. priciest things I ever worked on were DTV transmitters that ran about $650k and before that a $110k network analyzer system. Oodles of $70k spectrum analyzers though.

  13. It’s impressive. Back in the 80’s @ school we had to make a motherboard with a 6502 and pheripherals @ 1 Mhz. When it was ready, testing began with oscilloscopes that were then allready 20 years and had ofcourse self made cables. We had to measure the 1Mhz, it was there, not clean at all and we were wondering how this 6502 could still see these as nice blocks. We were impressed, this one goes 10000 times further, again, impressive.

  14. I will never buy anything from keysight, because their carelessness destroyed the HP archives. They moved the archives from a secure facility to a warehouse that burned to the ground.

    google “keysight hewlett packard archives”. two example articles
    https://www.pressdemocrat.com/business/7559762-181/hewlett-packard-archives-at-keysight-destroyed
    https://spectrum.ieee.org/view-from-the-valley/tech-history/silicon-revolution/loss-of-hewlettpackard-archive-a-wakeup-call-for-computer-historians

    bastards.

Leave a Reply to ShahriarCancel reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.