When I got the V2 user terminal (“Dishy”), I was lucky in that a few folks had already torn them apart: Ken Keiter went deep with a destructive teardown that shed a lot of light on the internals, and MikeOnSpace took a slightly less destructive (yet still power-tool driven) approach.
I couldn’t resist the opportunity to dig in a bit myself, and started collaborating with Lennert Wouters on the disassembly and EMMC dumping. Lennert went much further – eventually posting an awesome writeup on what he found.
So, having had a peek into V2, I was anxious to see what changed on V3. Were they using the same components – SoC, antenna setup, motors, etc? Or had the whole design changed with the updated form factor?
Don’t care about how the sausage is made? Skip to the PCB.
Enclosure – V2
First, some background on V2.
The V2 uterm was designed to be somewhat modular, consisting of a metal base and the user terminal itself, which could in turn be disassembled into two halves – the mount/motor assembly and the antenna/PCB assembly.
On V2, it’s possible and relatively easy to separate these halves. Once the back motor cover is removed, four torx bolts securing the housing to the PCB/antenna assembly can be removed and then it’s a simple matter of releasing (with a wedge or pry tool) the captive clips that run along the perimeter of the PCB assembly:
That was the easy part. Getting the heatsink/backer plate off the PCB was much harder, requiring careful application of a heatgun and pry tools:
I had expected a similar design with V3, but the design turned out to be much different.
Enclosure – V3
Unlike V2, the V3 uterm did not have any obvious seams, nor did it have a removable cover in the back over the motor compartment. Already off to a bad start!
The only familiar part seemed to be the rotating cover at the hinge point, that had enough flex in it to get a borescope in. I don’t have any pictures, but it was clear that the PCB wasn’t waiting just beyond that cover, and that there was some kind of motor compartment that seemed to be sealed off from everything else.
As a next step, I was able to pry the connector/cover inside the bottom of the post out, which opened up the post enough to get the borescope up it. This shed a bit more light (literally) on what the inside looked like.
…which didn’t help much at all. It was pretty clear that the motors and related parts were inside their own compartment with no obvious way to get to where the PCB was. Time to get a bit more aggressive.
Each side of the enclosure has two sets of three small holes that appear to serve as vents. Since this could allow water in, I figured they must vent to the motor area, not the actual PCB area. Seemed like a good place to start cutting.
Got out the xacto knife and started trying to cut out a larger area.
That didn’t quite go to plan – the blade broke quickly. I tried to pick the blade out with pliers, and found that the housing itself was brittle enough that it snapped.
It’s a weird material – it breaks almost like ceramic – leaving a dusty irregular fracture. I later found markings that indicate it is “PC+SAN” – polycarbonate plus styrene-acrylonitrile, apparently.
In any case, this is the point where it became clear this was going to be a destructive teardown, so I just kept going with the pliers 😬
A few minutes later, and I had a big enough hole to get the borescope in. Turned out my suspicions were correct, and there was in fact a separate motor compartment, sealed from everything else.
So that’s not much help. But as luck would have it, I ended up right at the corner of where the inner “tray” met the top housing, which meant I could pry away some more of the enclosure and (hopefully) get the borescope into the PCB area.
We’re in! Some immediate reactions:
- no heatsinks! None of the beamformer or RF chips have any thermal relief like they did on V2
- those RF chips look familiar…
- that motor connector looks familiar…
- it’s totally hollow, so we can probably cut all the way around the perimeter
OK, now we have a plan.
I have to say, this part was pretty crude. I’m not proud of it, but ultimately it worked. Basically, I kept breaking off pieces of the hole I originally started with until I had a big enough area to get a good look and work the borescope around to make sure there wouldn’t be anything in the way of a cut around the perimeter.
It all looked good, so…. time for power tools.
Some quality time with the Dremel and we’re almost there:
That got us most of the way there. I ended up using an oscillating saw to carefully cut down the “walls” and get to a flat PCB that was fully exposed.
As I picked away at it, I still couldn’t tell how the back enclosure was mated to the PCB/antenna housing – until a big piece broke off intact:
It turns out the back enclosure has a tongue on it which sits inside a groove on the PCB assembly. What’s interesting is that it’s not glued – or at least, not just glued, as far as I can tell. In between the tongue and groove there is some kind of matting or fabric that is about 1cm wide and gets sandwiched between the two halves, almost like a gasket. It’s possible this “gasket” has adhesive or some kind of glue applied to it, then the two halves are press-fit together.
Whatever it is, it doesn’t come apart easily – it seems like it chemically mates the two halves together, as in many cases the pieces broke outside of the mating surfaces.
The motor compartment cover was attached in the same way:
With everything apart, it was now possible to get a better look at the PCB itself. I was very careful not to damage it, so that it might stay usable.
Speaking of the PCB, let’s get to it…
So, here it is, the V3 PCB:
Ultimately, there aren’t many surprises here. It seems like V3 uses almost all of the same components that were present in V2 and prior.
Here you can see the main components:
- System-on-Chip (SoC) and EMMC, which are underneath the black heatsink
- Motor and ethernet connectors, which appear to be the same as V2
- Clock, same as V2
- Beamfomers, same as V2 (
GLLBSUABBBA, which seem to be custom for SpaceX)
- RF chips (amplifiers?)
- GPS (likely
ST Micro STA8089)
I have not yet removed the heatsink, but I do not expect any significant changes from V2 – it is almost certainly the same SoC based on other research.
The board has a marking in the solder mask that reads “COMPEQ”, which appears to be a PCB fab SpaceX imports from. It’s safe to assume that this is who makes the PCB, but maybe not the whole assembly.
The front stackup appears to consists of multiple layers of different material:
It’s tough to tell what is what, but it appears there are at least 5 layers:
|Front plastic (black) with white vinyl(?) membrane on top|
|Inner white plastic layer that appears to have antenna traces|
|Thin black membrane (middle)|
What’s interesting is that unlike V2, these layers don’t appear to be permanently bonded together, rather they are loosely sandwiched and held in place by numerous plastic (?) rivets throughout the entire PCB:
I suspect that by breaking these rivets off, it would be possible to separate all the layers above. However, it seems like at least some of the layers are integral to the RF magic, so I’m not going to do that now.
To fully appreciate the difference in PCB size, here’s V2 vs V3:
It’s been on a diet, as well. The PCB assembly comes in at just over 2.5lbs, or a little over 1kg.
That’s a significant reduction from V2, which weighed in at just about 4lb (1.8kg) without the metal backing.
So what’s different?
For one, the PCB is obviously smaller. It’s also rectangular, which likely means it’s easier to produce at scale. Although it’s not completely square (it has rounded edges) the reduced size and rectangular shape likely reduces the manufacturing complexity.
There are 16 beamformers on V3, compared to 20 on V2, and 80 on V1. There must be some new efficiencies here, but I’m not sure what. RF is black magic as far as I’m concerned.
V2 had trace heaters onboard, V3 uses only heat generated by the beamformers.
To what end?
So, why rip apart a brand new uterm?
Mostly because I’m curious. I find it incredibly impressive what SpaceX has been able to accomplish with Starlink (and pretty much everything else they’ve done) and it’s fascinating to see how this technology comes together. It also enables further exploration within the confines of SpaceX’s published bug bounty program.
The good news is, no Dishys were killed in the processed. Harmed, maybe, but I plugged it in shortly after completing the teardown and it still works. Which is good, because there’s more exploration to come 😎. I’ll post back here as it happens.