18

Today I stumbled across a piece of information that surprised me. Apparently there is a minimum cable length for SATA. From Seagate, a reputable manufacturer of SATA devices:

Serial ATA (SATA) data cable lengths

Serial ATA cables are available in many lengths up to 1 meter. Minimum cable length is 12 inches, using shorter cables can cause timing, or noise interference on the cable.

That is all the information they offer, and I could not find much information elsewhere. It is obvious that a cable can be too long: Signals will arrive too late, and eventually become attenuated. This is the first time I hear about cables being too short.

Since they measure the minimum length in inches it makes me think that it is not in the official standard, and something that they have found by trial-and-error, but can someone explain why this would be a problem in a (presumably well designed) standardized high frequency serial protocol?

Voltage Spike
  • 75,799
  • 36
  • 80
  • 208
pipe
  • 13,748
  • 5
  • 42
  • 72
  • 2
    I dunno but all the SATA cables I have ever bought and used are shorter than 12" – DKNguyen Jun 18 '20 at 18:55
  • 2
    Interesting - perhaps due to setup/hold violations from a reflection that comes back too quickly? Clutching at straws a bit..! – awjlogan Jun 18 '20 at 19:09
  • 8
    Laptops and tiny PCs like Intel NUC have much shorter SATA cables, or even no cable at all as the SSD plugs into the PCB directly.... Also SATA M.2 SSDs don't use a cable at all. So this is quite weird. – bobflux Jun 18 '20 at 19:26
  • 2
    Strange one. I've just built a PC and put 0.2m SATA cables in, they're freely available. – TonyM Jun 18 '20 at 19:28
  • 3
    I'm reserving judgement. However, with "Seagate, a reputable manufacturer of SATA devices", it seems that you're supplying evidence of the opposite. ;-) – Chris Knudsen Jun 18 '20 at 19:32
  • Two good answers to the contrary of the Seagate article. Maybe it was written in haste by a field engineer frustrated with bad cables and replacing with long cables worked, not realizing that the long cables were just better quality. – pipe Jun 18 '20 at 19:59
  • @ChrisKnudsen Hehe well, they _are_ a member of the SATA board so the literally "wrote the standard". Doesn't mean everyone who wrote stuff on their website knows about it though... – pipe Jun 18 '20 at 20:02
  • https://superuser.com/questions/1266468/can-a-sata-cable-be-too-short –  Jun 18 '20 at 20:16
  • @JonRB Thanks, missed that when searching, though I think it's still good here to find a more engineering-based answer. – pipe Jun 18 '20 at 20:47
  • agreed. it will be a signal integrity thing and thus the minimum length will be dependant on the specific manufacturer. I wouldn't be surprised those that state a lower short also have a lower long –  Jun 18 '20 at 22:27
  • @VoltageSpike the title edit doesn't make sense; OP knows there is a limit and indicates that very clearly in the body. OP's question is "why", and the title should reflect that. – TylerH Jun 19 '20 at 08:34
  • @TylerH OP is skeptical about the claim though, as is everyone else. – DKNguyen Jun 19 '20 at 13:29
  • @TylerH I'm perfectly fine with the edit, otherwise I would have reverted it in haste and written a snarky passive-aggressive comment about it. – pipe Jun 19 '20 at 19:37

2 Answers2

27

I have designed SATA asic serdes circuits and do not recall issues with short cables.

However, we do test for very long and very short length test cases to ensure sufficient signal integrity at either extreme, as sometimes a timing circuit or equalizer misbehaves when the signal is too clean.

For instance the timing recovery relies on a sloped transition between 0 and 1, and if the cable is too short, then the 0-to-1 and 1-to-0 slopes are too steep, and the timing recovery can become noisy (it's a long story). This can be an issue in asic test set-ups, but hardly in real deployments.

However, 12in seems to be a ridiculously high minimum.

P2000
  • 3,087
  • 7
  • 22
  • 2
    Could it be about mismatch at the connectors causing reflections that if the cable is too short could make a glitch appear within the hold time of the serdes? – Dan Mills Jun 19 '20 at 10:53
  • 1
    @DanMills, it's hard to say. I think an earlier commenter is right, that perhaps the short cables had poor mismatch, the long ones had better match (for reasons not related to the length), and so the FA (field application engineer) wrote a quick comment based on that observation, understandably, without further investigation. At 6Gbps I can't see that somehow a mismatch at 13in is worse than a mismatch at 4in. But again, from a distance, it's hard to say. – P2000 Jun 19 '20 at 16:46
  • 2
    I could see the reflected power from the mismatch making an edge non monotonic which might be a bigger issue for the CDR at the shorter length, the front end likely has way more then 3GHz bandwidth. I normally play with SDI rather then SATA but the rates are broadly similar. High speed signalling has some really funky ways to go sideways (And of course FAEs writing notes without actually investigating is no great surprise, one does not always discover genius). – Dan Mills Jun 19 '20 at 17:20
  • Out of curiosity, in your experience what length do the issues come from? I'm not familiar with what "asic test set-ups" are and I think just a small note about how long that implies would be super helpful. :) – Captain Man Jun 19 '20 at 20:36
  • @CaptainMan the set-up could be a loop-back out-to-in of the same ASIC that has two serdes connections -perhaps 1 inch loopback trace-, or between two ASICs generally on the same board, so they are perhaps a few inches apart. This is used to evaluate or test a pre-production spin of an asic, to perform S/W development at-speed, etc.. The serdes design itself might be subject to tests, or it could just be used as an IO to test other parts . – P2000 Jun 20 '20 at 00:25
  • The issue with too short cables would usually be reflected far-end crosstalk. But every specification I know of in the past twenty years has required the system to tolerate it. So unless you're using non-compliant hardware that can't handle the required level of fext, you should be fine. – David Schwartz Jun 21 '20 at 04:51
14

I searched the SATA 3.0 spec and could not find anything that would indicate there is a minimum cable length for SATA cables. I assume there isn't one.

There is a lot on the cables being too long, depending on the application and speed 1m to 2m seems to be the max length.

I would say design SATA cables as short as you want to. Seagate has made USB storage drives that have the SATA controller less than an inch or two from the controller. I don't see why they have included a minimum length on their website, it does not violate SATA specs (as far as I can tell).

I also can't think of a physical reason besides reflection that would dictate a minimum cable length, but the connectors are tightly matched so this shouldn't be much of a problem either.

Voltage Spike
  • 75,799
  • 36
  • 80
  • 208