What would be the effect of adding small (~<10mm) lengths of conductor between a CPU's contacts and socket

Question

[I am interested specifically in the case of state of-the-art consumer processors and motherboards]

To clarify what the setup might be in practice, imagine wires running vertically between each pair of contacts on the CPU and socket.

I don't have much expertise in this area, but these are the potential issues I am concerned about:

• The additional length of the conduction path from the processor slows down the speed of processing.
• The change in length of the conduction path means that the processor won't work at all as the devices are designed to operate with a specific length of conduction path.
• There is interference of some sort between the conductors running from the CPU to the socket.

In each case I am wondering if this is a problem and, if possible, what sort of effects would be observed.

Answer 1

Your CPU has several types of pins/signals:

• Slow signals like SMBUS, etc.

These wouldn't care.

• Fast signals like PCI-Express, RAM, etc.

These all use transmission lines, single ended or differential. Any impedance discontinuity in the transmission line would create signal reflections and corrupt signal integrity, so you would need to make a riser connector with controlled impedance. This can be done (and it is done) but it will involve tiny high precision machined parts and obscene prices because this is a very low-volume, specialty custom product. Not to mention how crosstalk between nearby signals would be handled. Perhaps tiny coax pogo pins? Add an extra zero to the price.

• Power/Ground

These require extremely low inductance which is achieved by having literally tons of pins. Here's a random PC CPU pinout from the internets:

Enlarge it and look at all the supply and ground pins (anything that starts with "V"). More than half the pins are supply and ground, and this adds to the cost.

Adding inductance (ie, length) would degrade power integrity.

These are the reasons why motherboard CPU sockets sit as low as possible on the board and keep the connections as short as possible...

So i would strongly recommend you find a cooling solution that doesn't require this. It is a lot easier to move heat up by a few mm (a 10 cents copper shim will do the job).

Answer 2

This would dramatically increase the inductance associated with each connection from the PWB to the chip. It would also increase cross-talk between data signals on those connections.

It wouldn't affect (much) the ability of the processor to continue to do computations at high rates.

But all interfaces to the outside world (for example to memory and any peripheral bus like PCI) would have to be slowed down considerably to be made to work. For example instead of accessing RAM at 200 million transactions per second you might have to slow it down to 10 million transactions per second or slower. And of course the existing memory chips aren't designed to work with those slow access rates.

Answer 3

If you can slow the output edges (output drivers) by 10:1, to perhaps 10nanoseconds, and you solder 0.1uF capacitors across all VDD/RTN pairs on this modern MCU, you might get away with this.