SATA 6Gbps - Docking connectors, signal integrity

[Tantalum]

Hello 

I’m building a PC Mod and I need to route some SATA 6Gbps (and maybe USB 3.0) traces through several connectors and I’m worrying about the signal integrity. Not because of the length of the traces, I think they won’t exceed 200mm, but I’m worry about the amount of connectors used.

Here you have a picture of the PC case (without the front panel)


You can see 3 stacked modules.
The upper one contains the PC motherboard and the SATA Raid card.
The 2 modules underneath contain the HDDs (2 or 3 each).
So, as you imagine I need to pass the 4-6 SATA 6Gbps connections from the upper to the 2 modules below, and therefore I wanted to use docking station like connectors.
Actually to give you all the details, that connector will carry those signals:
-   12V power to power up the HDD, and maybe the front panel, I’m not such if I will use a different power rail.
-   4-6x SATA 6Gbps
-   4-6x HDD activity signals (to connected to the front panel, well that’s another story....)
-   EN signal to enable the 5V switched regulators for the HDDs  (when the PC boots up)
-   I2C (connected to the  front panel MCU) + 3V Power
-   USB 2.0 + 5V (to connect the front panel MCU to the PC)
-   Maybe 2x USB 3.0  (I want to derive 2 connectors to the lowest module back panel)

The problem is the high speed buses (SATA and USB 3.0), all the other stuff is ok. How do I plan to do that? Well that way:


-   The female docking connector of the first upper module is connected through SATA cables to the SATA controller card.
-   On the 2nd module, 2 SATA connections are derived over a FFC connector and cable to a PCB where the HDDs are connected.
-   The unused SATA connections are derived over a FFC connector and cables to the female docking connector PCB of the 2nd module to feed the 3 module.


I also planned to add SATA redrivers if needed, like the SN75LVCP601 (datasheet) from TI  to reboost the signal, at each receiver side of the links (just before the signals enter the HDD and SATA controller card), or after each connector (like drawn on the picture. Only the downstream links are displayed, actually the upstream links would have another redriver just before the signal enters the SATA RAID Card).

I wanted to use JAE WD2 Serie  (site) (datasheet) (or similar if you know other ones!) docking connectors,  that are compatible with PCIe differential signals (similar to SATA).
But they don’t mention which maximal speed is supported.
The maximal impedance is also quite high (190mOhm), the SATA requires 100mOhm +/-15%.


Well I suppose it won’t work, but I really would appreciate opinions and suggestions of people who work with high speed buses.

Thank you!

edit:
(Sorry for my bad English  )

I have added the SATA redrivers on the picture.

[John_ITIC]

If you really need to ensure that the final solution will work and that it meets the specs with regards to error rate and signal integrity then you will need to either do a differential mixed-mode S-parameter analysis of the complete signal path. This will tell you whether the transmission path will work at all.

You could start by slapping the pieces together and use a BERT to see what the actual error rate is. You could also use a signal source (generator) and a scope with proper software for analyzing the eye diagram on the receiving side. Build some test jigs to see what the final characteristics will be.

For these type of buses, you really need to know what you are doing or things will simply not work. The margin for error is very slim.

[Tantalum]

Hi,

Thank you for your answer 

If you really need to ensure that the final solution will work

Well actually, it's not a commercial product, but yes I need to know if it could work, because this docking port solution implies such structural changes to the case that I could not finalize the design. 

If you really need to ensure that the final solution will work and that it meets the specs with regards to error rate and signal integrity then you will need to either do a differential mixed-mode S-parameter analysis of the complete signal path. This will tell you whether the transmission path will work at all.

The problem is that I doubt that Highpoint/Marvell (manufacturer of the RAID card/controller) will provide me those information. Same for the HDD controller manufacturer :/.
And I'm quite sure I do not have the knowledge to do such an analysis. Yeah I'm still a noob in electronics. 

You could start by slapping the pieces together and use a BERT to see what the actual error rate is. You could also use a signal source (generator) and a scope with proper software for analyzing the eye diagram on the receiving side. Build some test jigs to see what the final characteristics will be.

For these type of buses, you really need to know what you are doing or things will simply not work. The margin for error is very slim.

But do you believe, or have the impression it could work with those components (dock connector, FFC cable,....)?
What about the redriver? Should I built a test PCB with and another one without redriver?

(Those connectors are quite expensive so if I could avoid to assemble a whole bunch of different PCBs it would be great )

[John_ITIC]

Initially, forget about the re-driver. Your problem will be reflections due to impedance-discontinuities in the connectors in the signal path. Re-drivers are intended to compensate for effects of long signal runs through PCBs and cables. Based on your drawing, doesn't look like too long transmission lines.

If you are a noob then all you can do is to slap the pieces together to see whether it will work. If you properly want to *engineer* the solution then you need to rent very expensive equipment to guarantee a proper functionality.

Essentially, modern high-speed, gigabit buses are land squarely into the microwave domain so needs special equipment and methods to work with. Take a peek at the below links:

https://www.google.com/?gws_rd=ssl#q=high-speed+channel+characterization

[wraper]

IMO you could use SATA cables with panel mount connectors instead. This would give much more predictable result.

[Tantalum]

Well using SATA cables + PCB with SATA connector on which the hdd will connect to, was actually the initial idea, yes, but it causes some troubles.
(Yes PCB. Those connectors on the picture won't fit in the case, the inner width is only 118mm, and the 2.5" hdd occupies already 100mm)

For technical reasons it would be really great if I could use docking connectors on the modules.


I think I will make some tests boards with those connectors and see what happen (would have preferred to avoid those costs)
If the HDDs are recognized by the system and that I run massive read/write tests during several days on those HDDs, do you think that I could rely on the SMART infos to verify if the connexion is really reliable?

[wraper]

2 days won't tell too much. I had a faulty SATA cable which caused Interface CRC Error count to grow by one once in a few months. Figured this out only when used this cable with different HDD on different port and got the same thing. Took like a year to finally trace the issue. Since then I just discard any cable is suspicious even by tiny bit.
You opened this tread in beginners section, so may I ask if you understand that need at least 4 layer PCB? This will make the project quiet expensive.

[wraper]

Also, even if this works OK. You may find that for some reason wireless mouse works awkward or wi-fi performs worse than expected. And this is just because your beards are HF interference transmitters.

[Tantalum]

You opened this tread in beginners section, so may I ask if you understand that need at least 4 layer PCB? This will make the project quiet expensive.

Are you sure?
That docking port has far more pins than I actually need (there is only a 144 pin version available, the 100 pin version is angled), so I think it will be possible to route all those traces on a 2 layer PCB, if not, well it will get expensive indeed.

Also, even if this works OK. You may find that for some reason wireless mouse works awkward or wi-fi performs worse than expected. And this is just because your beards are HF interference transmitters.

Well every module is built in aluminum, the small vent holes use aluminum mesh and the docking connectors are shielded, so I think it will be ok with the EMI (Well, lets say I hope so ).





There is an alternative possible which reduce the number of connectors.

• The SATA signals go only through the first upper Docking port.
• The SATA Raid card is connected to the female docking port PCB with a standard SATA cable.
• The male docking port PCB on module 2 has 4 standard SATA connectors
• SATA cables connect those ports to the final PCB on which the HDDs are connected
• Module 2 & 3 will still have there own docking port to connect each other, but it will only route power rails and low speed signals. (Not drawn on picture)

That version has far more chances to work. Not?
(One docking port, no FFC unshielded cable)


Now I have 2 questions.

1. As you see on the picture, the SATA cables have only a connector on one side. The other end is soldered on the PCB (on surface, not through hole). Is soldering the cables to the PCB a better solution than adding another SATA connector?
(I had to cut the cables anyway, because I won't find cables with the exact length needed)

2. There is another solution to reduce the cable mess. On the module 2, I could connect both PCBs with pin headers. Is that ok?
 

There are cards available that use similar pin headers to route the SATA signals to internal ports or to external eSATA ports, so I assume that those pin headers are ok for that purpose. Not?





Thanks again for your opinions, I appreciate your help.

[T3sl4co1l]

Pin headers can be used in a controlled-impedance fashion.  Sometimes, the datasheet will give an impedance rating for a given configuration, usually alternating signal and ground.  Use as many ground/VCC pins as possible (at minimum, at least as many GND/VCC as signal pins), to provide shielding for the signals.  The high-speed board-to-board connectors handle this by using paired and grounded/shielded configurations.

I don't know that SATA cable can even be soldered to a PCB.  I can't say I've ever seen such construction.  It seems likely to me, the internal shield will be aluminum foil, which cannot be soldered, and that's a problem.  Shielding is required for all these signals.

If you're that desperate, it might be worth going out of your way to get/make some flex (FFC/FPC) material with the required characteristics. 

FYI, PCIe can also be extended, since the data lanes are similar to SATA lanes.  I forget what all control signals accompany those, though.

Tim

[wraper]

You opened this tread in beginners section, so may I ask if you understand that need at least 4 layer PCB? This will make the project quiet expensive.

Are you sure?
That docking port has far more pins than I actually need (there is only a 144 pin version available, the 100 pin version is angled), so I think it will be possible to route all those traces on a 2 layer PCB, if not, well it will get expensive indeed.

This is not about pin count but signal integrity. You won't be able to maintain needed impedance on 2 layer board.

[Tantalum]

It seems likely to me, the internal shield will be aluminum foil, which cannot be soldered, and that's a problem.  Shielding is required for all these signals.

I also asked myself that question.


Actually the foil is connected to... nothing, since the connector is made of plastic.
But that foil is in contact with the 3 ground wires (Drain), so I think, I may be wrong, if those 3 ground wires are also connected to the PCB (as they are usually over the SATA connector), the shielding will still work effectively. Not?

If you're that desperate, it might be worth going out of your way to get/make some flex (FFC/FPC) material with the required characteristics. 

Probably far too expensive 

Are you sure?
That docking port has far more pins than I actually need (there is only a 144 pin version available, the 100 pin version is angled), so I think it will be possible to route all those traces on a 2 layer PCB, if not, well it will get expensive indeed.

This is not about pin count but signal integrity. You won't be able to maintain needed impedance on 2 layer board.

Because of the dielectric thickness?

[wraper]

Because of the dielectric thickness?

Yes, too thick. Also if you split the ground pour on the opposite side by running some tracks on it, signal integrity will be destroyed because of this too.

[Tantalum]

Thanks for the information.

Well, I will think about that and take a decision  .