A Step Towards Liquid Cooling #Heat_SinkDecember 6, 2018
Most chips cool in ambient air that gets blown through by a system fan.
So this provides a way for a chip to exchange cooling liquid with an interposer – or with another chip, in a 3D stack.
Coupling between Processor and Memory
Because they were using standard chips, they kept the heat sink in the interposer rather than relying on a change in the dice to make this work.
In the standard configuration, which puts a heat spreader over the two chips and a metal heat sink on the processor, the simulation heats the processor up to 102 °C, while the memory gets up to 86 °C under their operating assumptions (which I’m omitting, but which you can find in the paper).
Cooled fluidically next to each other, without the heat spreader and metal heat sink, the temperatures dropped to 63 and 46 °C, respectively, showing not just better cooling, but also reduced thermal crosstalk.
While this might seem silly, since this keeps the processor away from cooling air, it again relies on the electrical connections to guide the heat down into the interposer, away from the memory.
Under the thermal bridge, a second microfluidic heat sink was used to draw this heat away.
The idea was to see how many processors they could run before the thing got too hot, comparing a standard air-cooled version (stock with the dev board) against their fluid-cooled hack.
Polymer vias for electrical connections in a thick interposer; Fluidic microbumps with vias for mating dice to dice or to interposers; Separate fluid circuits for separate chips to reduce thermal crosstalk; A new 3D stacking architecture for processor and memory, including an air gap between stacked dice, die-to-die TSVs clustered away from sensitive circuits, and a thermal bridge with its own separate heat sink in the interposer; and A way of creating fluid heat sinks by etching a “pin forest” (my term, not theirs) into the backside of a wafer.
Usually a heat sink is attached to the integrated heat spreader (IHS), essentially a large, flat plate attached to the CPU, with conduction paste layered between. This dissipates or spreads the heat locally. Unlike a heat sink, a spreader is meant to redistribute heat, not to remove it. In addition, the IHS protects the fragile CPU.