Why the push towards serial?

[Amorphus]

these days, everything fast is using serial technology, and its replacing a parallel technology. i.e.
Serial ATA (serial data transport) - Traditional ATA (parallel data transport)
USB/Firewire (serial) - Parallel port (parallel... duh).

and so on.

well, theoretically, since those are serial, wouldn't greater speeds be attainable by using improved parallel technology? for example, two USB ports doing the work of one would provide almost a gigabit of bandwidth, (and we would have to call it UPB... =/ ). so why are we sticking with serial? it seems to me that one of these days, the switch to parallel would be made again.

 

[spidey07]

Possibly due to the success of network technologies that are all serial? Plus it seems like a simplier design from a connector, logic and clocking standpoint.

 

[Lynx516]

It is cheaper and easyer to implement.
It is much easyer to put 8 traces on the board and have a clock at say 400mhz then to have 64 traces and have the clock at 50mhz. It is a question of space. More traces mean more space is needed and more space means more PCB layers and more layers means more money.

That sums it up pretty well

 

[Peter]

Easier to implement is not exactly right. OK, so its much fewer traces, but the signal frequencies are massively higher (we're talking 2 GHz and more here). This requires a lot more care and babysitting in board layout and routing.

But the fewer signal lines is where the actual gain lies - it means smaller chip packages, shorter card connectors, lighter cabling.

So why now, not earlier? The usual two factors ... one, science has advanced enough that the new technology becomes manageable; two, the old technology has become a pain in the arse.

 

[TDSLB]

Just a thought, if a bus had to run with, like peter said, 2 Ghz lower frequency then it's Serial 'conterpart.' Wouldn't this also conserve energy and therefore make computers overly more efficent for having to move less electons? or am I mistaken? It just seems like the fact we have motherboards that run at such inefficient levels that they have to use fans to cool wouldn't it make more sense to use a parrell bus? or am I just lost and should be flogged?

 

[CTho9305]

Originally posted by: Peter
Easier to implement is not exactly right. OK, so its much fewer traces, but the signal frequencies are massively higher (we're talking 2 GHz and more here). This requires a lot more care and babysitting in board layout and routing.

But the fewer signal lines is where the actual gain lies - it means smaller chip packages, shorter card connectors, lighter cabling.

So why now, not earlier? The usual two factors ... one, science has advanced enough that the new technology becomes manageable; two, the old technology has become a pain in the arse.

It is difficult to keep many wires synchronized (at high clocks / over long distances) - with the serial case you don't have to worry about that.

 

[sciencewhiz]

serial and parallel technologies seem to go in cycles. Eventually today's serial technologies will get even cheaper, and hit a speed wall. Then someone will say "hey, we could put 8 of these lines together and be 8 times as fast". Then parallel will make a comback. Then speeds will improve again, and it will come to the point where it is again better to do serial, and so on and so forth.

 

[Peter]

That'll be multi-strand serial then, not really parallel. As CTho9305 correctly explained my "pain in the arse" statement, getting parallel busses layed out and routed correctly gets increasingly difficult with number of data lines and frequency. With higher frequency, issues like trace length differences, wire impedance, crosstalk and other details become more important; and this is exactly what becomes increasingly impossible when you make the bus wider.

 

[RossGr]

Seems to me that with modern 32 and even 64 bit busses a 8 bit parallel line is not sufficient, shouldn't be going to a wider "parallel" port. I would guess that it is pretty much the same process to parcel out a 64bit buss on to a 8bit line as it is to go completely serial. In that respect serial would be compatable with all future busses. Why would we want to continue to use the BYTE sized buss when we are dealling with much longer WORDS?

 

[AEB]

i dont like what they call it, when i think serial i think of the old serial the one where data travels bit by bit in a 9 bit string, but does it matter what they call it as long as they keep improving?! i haven't baught serial ATA yet but USB 2.0 comes in very handy espcially now with the 1.0 GB thumb drives.

 

[dangereuxjeux]

Having looked at Serial ATA quite a bit, a few advantages:
transfers at voltage of 250 mV rather than the 5V of Ultra ATA
elimination of signal interference (two phase-reversed signals)
cables have less wires (7) and can be much longer (1m)
can be sped up to 300MB/s and then 600 eventually (according to roadmaps)

*all these brought about by this simple problem with PATA - parallel signals work fine at lower speeds, but as they have been ramped up more and more and speeds increase, it becomes increasingly difficult for signals to synchronize and interference to be kept in check ( interference was the reason for the additional 40 pins added for UDMA/66+) but SATA obviously avoids this problem by having only one signal sent much faster

 

[Moohooya]

I don't think this will save much energy. The faster the bus, the more energy it will use. Even though you cut down the number of signals, you'll still use the same enery in the hight bus speed.

However, that said, the lower the voltage, the less energy used, and the lower the heat dissapation. Serial seems to be using lower voltages. I'm not sure if this is just because it is a newer technology, or if parallel requires higher voltages to try and keep the signals in sync.

Originally posted by: TDSLB
Just a thought, if a bus had to run with, like peter said, 2 Ghz lower frequency then it's Serial 'conterpart.' Wouldn't this also conserve energy and therefore make computers overly more efficent for having to move less electons? or am I mistaken? It just seems like the fact we have motherboards that run at such inefficient levels that they have to use fans to cool wouldn't it make more sense to use a parrell bus? or am I just lost and should be flogged?

 

[wviperw]

Okay, we're talking all about yesterday's and today's methods of data transfer, but what is tomorrow's? What is the logical endpoint where this will end up being at? Serial or Parallel? It's almost like RISC vs. CISC in a way, and it seems that area is heading towards simple as well.

 

[spidey07]

Originally posted by: wviperw
Okay, we're talking all about yesterday's and today's methods of data transfer, but what is tomorrow's? What is the logical endpoint where this will end up being at? Serial or Parallel? It's almost like RISC vs. CISC in a way, and it seems that area is heading towards simple as well.

Well from a network perspective it will most likely always be serial. 10 Gigabit is out, 1 gigabit has commodity pricing, and if it follows previous trends 100 Gigabit will be out in 6 years or so.

Seems to me given the simplicity from a cable perspective, logic and speed...serial wins.

 

[kpb]

Parallel and serial have different advantages and disadvantages.

At low speeds parallel is easier to get going faster. The issues of cross talk and such aren't as much of a problem and the actual electronics run at a slower speed which made it the perfect match for the older computers. Your not going to get ghz buses with the processor running at 200 mhz =)

Serial on the other hand requires much fasters signaling speeds to get the same transfer rate but doesn't have as much trouble running at those higher speeds since crosstalk and other issues are minimal. It also makes nice small cables and uses less traces/connectors on mb's and such. This pairs up well with modern tech since we can now make chips running at much higher speeds making ghz serial busses possible.

as peter said things are more likely to go multilink serial. Infact most things already have that. The difference being that instead of of reading it across the lines 1 bit at a time it's broken up into 8 bit (or what ever) chunks and sent just like serial except it sends 2+ chucks across the multiple links, 1 per link and on the other end you just need to put them back in the right order. This prevents having to exactly sync the multiple lines together.

 

[Shalmanese]

What exactly IS crosstalk and why is it such a problem? Does it become inherently a problem at X speed or is it something that can be overcome with bigger and better technologies? What are the disadvantages of multi-link serial as opposed to parralel?

 

[CTho9305]

Crosstalk is when the signal on one wire affects another wire... it has to do with changing magnetic fields, and higher-speed lines change faster, so tehy have a bigger effect on other wires. If I remembered more high school physics, I could probably explain it better

 

[Shalmanese]

So... If we had a 500Mhz signal running at 5V, it would produce roughly the same amount of crosstalk as a 1Ghz signal running at 2.5V? So the limitation of parralel is really how low we can drop the voltage before the signal to noise ratio goes up.

 

[Lynx516]

Dont think it works exactly like that but you kinda get the idea. The higher the voltage change the more voltage is induced in other wires. Also the speed of the change will do the same thing. The faster it is teh more voltage is induced

 

[sgtroyer]

Shalmanese, you're missing an essential point. Reducing the voltage by half decreases the crosstalk, but it increases the susceptibility to crosstalk. Suppose a 5V signal induces a 500mV glitch on an adjacent trace. That's 10% of 5V or an SNR of 10. Now reduce your voltage to 2.5V. The noise glitch is now only 250mV, but it's still 10% of 2.5V. The SNR remains the same. Reducing voltage swing helps with power dissipation, and is necessary for modern processes, but it doesn't help with crosstalk.

 

[Peter]

That's why any serious cabling uses differential signalling. By encoding the data as the difference between the voltages on a mirror pair of signal lines, the usual forms of crosstalk (which affects both lines in the same direction) does not disturb the data.

This is one of the reasons why it's pretty urgent to get rid of that crap parallel ATA cabling - it's a technology originally made for 2 MB/s, still using high voltage single-ended signalling for 133 MB/s now ... the ever shorter cable length limit and the extra ground wires in the 80-wire cables are just duct tape; SATA is a solid new solution to getting the data there.

 

[borealiss]

Originally posted by: Shalmanese
What exactly IS crosstalk and why is it such a problem? Does it become inherently a problem at X speed or is it something that can be overcome with bigger and better technologies? What are the disadvantages of multi-link serial as opposed to parralel?

crosstalk has to do with the inherent conductance of a signal traveling along a wire. similar to inductance transformers.

Originally posted by: Shalmanese
So... If we had a 500Mhz signal running at 5V, it would produce roughly the same amount of crosstalk as a 1Ghz signal running at 2.5V? So the limitation of parralel is really how low we can drop the voltage before the signal to noise ratio goes up.

also remember that newer technologies in signaling have made faster signaling possible. usually traditional parallel buses have used conventional signaling, ie 1 trace per bit, but technologies such as hypertransport use differential signaling. so things like noise become less of an issue because the difference between the signal and its complement determine logic high or low.