Grab a chair, sit down, keep hanky at bay because we are entering eye-watering territory.
It's not exactly a full teardown but this is as close as you'll ever get to this 'Unobtainium' Scope.
The short specs :
- 33Ghz analog bandwidth ,
- 80 gigasamples per second
- 2.1 Billion points of trace memory.
Cost ? More than your average house. Base model starts at 300K$ and a full blown all options tops well over half a million dollar.
The Beast
For the non-believers : Yes we can ...
This machine was under development for many years and the designers were in uncharted waters. They tapped into the know how of Agilent's RF division and even those guys were scratching their head. Never been done before ...
Silicon germanium ? too noisy. Gallium Arsenide ? Too slow .. Indium Phosfide then ? Ah, but our semiconductor fab in Santa Rosa can run that ... Yes boys and girls. This machine is fully built in USA. While most of the scopes are assembled in Agilent's Malaysia operations, this beastie is not. It was designed by Agilent's Scope operations in Colorado , the Santa Clara people did the digital guts and the Santa rosa operation made the analog front-end and A/D converter.
The input :
The high bandwidth meant that they had to let go of the classic BNC or even standard SMA connectors as even those top out at 10 to 12Ghz... Instead Agilent opted for precision 3.5 mm connectors with built-in torque control. Twist the big grey ring and when it 'clicks' it means you now have applied sufficient torque to guarantee correct and repetitive performance. Besides the input connector there is another multipin connector that provides power and communication with the active probe heads. The active probes store s-parameters which the scope reads and uses to compensate the signal pathway, effectively shifting the plane of measurement to the probe tip and not the machine input.
Even then, the accuracy may not be enough for what you want to do so the scope has a built in calibrator output .
This delivers a 15pS rise time pulse pattern ( long dead time active high , followed by a low-high-low-high transient ) the scope uses as a kind of TDR measurement to compensate it's signal path further.
The 3.5mm connector is almost SMA compatible. Even though they will mate , long term damage may be inflicted and Agilent has so called 'port-savers' that you can screw into the connectors. The input will take a 5 volt pp signal without blinking. Beyond that you need attenuators.
Needless to say the cables and attenuators cost a fair amount of money as well... the little attenuator above costs 438$... the attached cable another 520$ ...
So what makes this thing tick ?
The magic happens in the front-end hybrid.
link to the full blown picture
http://www.siliconvalleygarage.com/eevblog/unobtainium/ad-ori.JPGThe chip closest to you is the A/D convertor. the chips furthest away are the input amplifiers ,The chips in the middle are (i assume) the triggering logic.
The A/D is fed power and control signals through the traces going sideways. This is relatively low-speed and parallel. The data is streaming out through 8 differential unidirectional serial buses. It's the only way to get the data out at this breakneck speed. This module is a cross breed between a hybrid ( the blue area ) and a BGA ( the green area. ) Chips are bonded directly to the substrate and the hybrid is bonded in turn to the bga area ( you can see the little solid gold bond-wires of you zoom in ).
view from the other side :
link to the full blown picture
http://www.siliconvalleygarage.com/eevblog/unobtainium/ad-front-ori.JPGthis shows the input amplifiers in close up as well as the input protection and connection.
Notice the wide trace coming down from the input amplifier , going through a 'tunnel' and ending, seemingly, nowhere ?
well, let's flip the chip over shall we :
There they are : the input connectors. The single ended signal could not be transported to the amplifier through the package. So they just bolted RF connectors on the chip's back... Voodoo-territory..
The A/D output buses are fed into another ASIC that demuxes and parallelises the streams into fast GDDR5 memory. Beyond that , it's just another infiniium scope.
I was able to use this machine to test the famed Jim Williams Pulse generator and record once-and-for-all it's rise time.
And here's me :
Yes, i can use this thing. It's no more complicated than a regular infiniium scope. Childs play. The only thing you need to be aware of is to run probe-compensation when you are getting ready to measure. Connect your measuring plane to the calibrator output , click the probe compensation menu and let the scope figure out what is the behavior of the stuff you attached. This effectively shifts the 'system input' to the tip of the cable.
Making measurement is drag-n-drop just like with any infinium. The user interface is extremely responsive and button pushes or encoder clicks deliver instantaneous screen update. As usual , everything is done by raw hardware and the built in computer only serves as user interface , file storage and remote operation. While it runs windows you don't notice it at all (unless you start saving screenshots etc , then the traditional windows explorer opens). It's main cpu isn't even that powerful.
The machine is silent enough you can leave it on your desk. It has 12 big fans forcing air over the digitizer board but since they used the whole side of the machine to convey the airflow there is virtually no fan noise.