At the physical layer, is it the materials in the cables are getting better? Or new ways of using the same materials?
At the protocol layer, is it newly developed computer science theory being applied or is it just old fashioned pragmatic engineering iteration, looking at usage and making the existing protocols more efficient with tricks and shortcuts etc?
So is it not actually serial any more?
Parallel refers to taking a raw bit stream and sending N bits at a time down N wires to a receiver who reads N bits at a time and reconstructs the bitstream. This was a common technique in the early days of computing because it is very simple to implement and low clock speeds meant it was often the only practical way of increasing the throughput of an interface.
As clock speeds increased, keeping each of the N signals in sync became increasingly difficult. This eventually lead to parallel interfaces falling out of favor and serial interfaces becoming dominant.
The defining characteristic of a serial interface is that the data signal itself also serves as the clock signal which keeps sender and receiver in sync.
Modern "parallel serial" interfaces like USB 3 consist of multiple serial links which are multiplexed using a framing protocol of some sort. So from a signaling standpoint they are still considered serial interfaces, even though data is in fact being sent in parallel.
-Power
-Ground
-Signal Ground
-USB2 data+
-USB2 data-
-USB3 RX+
-USB3 RX-
-USB3 TX+
-USB3 TX-
AFAIK, it's not, the SuperSpeed (USB3) wires have their own independent negotiation; the device switches to the USB2 wires only when USB3 is not detected. This also explains how the proposed VirtualLink alternate mode can have USB3 without the USB2 pair.
edit: yikes, looks like USB C doubled the number of those separate pairs, for a total of 6 data pins (2x RX, 2x TX, 2x USB 2.0 RX/TX)
Yes, and more expensive on that. Laptop makers totally hate usb 3.0+ for that.
The industry standard for internal digital connectivity was the FPC - flexible printed circuit board, a very economical and ergonomic solution, but then came USB 3.0.
Regular polyimide FPC can't reliably handle the high frequency signalling. Still, many laptop makers still use it, to regrettable results.
The few solutions left are the micro coax cable, running twisted pair inside the case, or passing it over a hard pcb bridge. All are very expensive, and hard on assembly lines.
This is why in most budget laptops manufacturers do a following trick – they put all usb 3.0 ports on one side, and 2.0 on the opposite. This way they only have to run the low frequency, low lane count usb 2.0 to other side of the laptop.
Only very recently, there came an FPC maker that makes custom cabling with some proprietary dielectric that can handle 3.0. And they charge more for it than for hard pcb bridge.
Even if the cable material issue is solved, you still have to find way to connect it to PCB. High frequency connectors are not cheap either.
Color me unconvinced on one point though:
> Only very recently, there came an FPC maker that makes custom cabling with some proprietary dielectric that can handle 3.0.
A brief glance at IPC-4203A (which has been around since 2013) tells me this is a consequence of market cost sensitivity, not one driven by the inherent performance limitations of available material.
The latest IPC-4203B was just published circa March 2018 last year. Is there something new worth looking into here, or is this "recent advancement" just a manufacturing process-driven one which has made high-performance FPC more viable in the throwaway consumer electronics market?
Each successive frequency increase has required more complex PHY design.
On the digital side : more sophisticated bit encoding schemes
On the analog side : tighter tolerances for clocking and signal jitter, more sophisticated transmitter and receiver technology (better emphasis/deemphasis, better equalization)
Here's taste of the PHY changes : https://www.chipestimate.com/Type-C-USB31-PHY-Challenges-in-...
USB-C has 4 differential twisted pair data lines.
USB 2 has a single twisted wire pair, specced for moderate speed.
USB 3 replaces that with two pairs with much tighter requirements, one for transmit and one for receive. This involves a tenfold speed boost (.5Gbps to 5Gbps), with much more modern transceivers using much higher frequencies on much better wires. But it also drops the length a single passive cable can be from about 5 to 2 meters.
USB C doubled the number of pins while making the plug reversible, so they went ahead and made the special case of C-to-C cables connect to all the pins and have four pairs of wires.
But that's only a doubling from 5Gbps to 10Gbps. How are we getting up to 20 or 40? It's mostly coming from restricting cable length even further, down to 1 meter or only half a meter. But on the bright side you can get an active cable, with chips inside, that can get you back up to 1 meter or 2 meters or 60 meters. As long as you're willing to pay enough.
I have a couple of situations I would like a 7m-ish Thunderbolt 3 cable but can’t seem to find one, even for the multiple-hundreds it would probably cost.
It's really disappointing that Display Stream Compression hasn't been built into everything for the last 5+ years, and is only now barely getting into equipment. 2:1 or 3:1 compressed display data looks fine, and being able to add more pixels or bit depth or refresh rate far more than makes up for the tiny losses. Without it, tons of screens are stuck at significantly reduced framerates or 4:2:0 subsampling.