Technology continues to advance at a fast pace: Ten years ago a 233MHz single-core processor which wasted as much energy in heat as it used for work was the considered norm. Floppy-drives were still the thing, and CD-ROM drives were fitted to computers as standard: CD-Rs were on the horizon. PCI graphics cards were in vogue, and AGP slots were being introduced on the latest motherboards. Microsoft were developing Direct X 7 and Windows 98  was becoming the most popular OS: We’ve certainly come a long way since then.

Processors have advanced in leaps and bounds since; with Intel pioneering Hyperthreading technology in the later models of the Pentium 4 range, AMD introducing the rise of dual-core processor technology with their Athlon 64s, then Intel rising as market leaders again by bettering AMD’s dual-core performance, followed by stapling two dual-core wafers together in a single package to produce the first quad-core processors… Despite the lag by software writers in utilising multi-core technology there are currently plans for eight, twelve, and sixteen-core processors; largely thanks to miniaturisation of the transistor from work pioneered by Intel since the 1990s: 32 nanometer transistors are now being built, 28 nanometer transistor production is on the horizon, and scientists have even succeeded in building a transistor with a single atom if obscure reports are to be believed.

So let’s look forward a few years: We see millions of single or dual-atom transistors being packed into advanced-level processor cores; 64 to a wafer, making up a unit only slightly larger than an Athlon 64×2 chip and with a few more pins. With advances in core architecture, core-interfacing, and cache-data-interfacing technologies these latest chips provide quite a punch in processing power and performance. Software is being written to allow from 2 to 128 core operation, and things have, in a way, hit the buffers: Transistors are as small as they can get; marginal performance gains are still being made by using new semiconductor-doping techniques to slightly improve the switching-speed of the millions of tiny transistor-arrays within each core, constructed across double gallium-arsenide molecules doped onto the silicon wafers; but other than that it’s a virtual stalemate: There’s no point in stapling wafers together, as since the component-miniaturisation limit has been reached, that’ll just lead to big packages which waste room on the motherboard – So where to go from there?

In this future-world there’s always the chance of multi-chip motherboards being manufactured: Motherboards with sockets for up to eight of these packages, but motherboard miniaturisation has almost reached critical mass too. Well when you’ve expanded outwards as far as you can the only way is up – Literally: Wafer upon wafer. The problem now is cooling. These latest packages, despite the huge technological advances, still produce as much heat as the old Athlon 64×2 chips did; and that’s with only a single wafer. Stacking wafers without an advanced cooling system would lead to conflagration and component failure.

The answer: Liquid cooling using high-pressure non-inflammable isobutane-based refrigerant running at low-velocity through tiny flattened spaces between wafers and in coolant-tubes between cache-interfaces within a solid-state cube-shaped chip with connection-pins on all but two sides: One of the unused sides is a tiny TFT monitor displaying real-time performance-statistics for the “cube-chip”; This monitor can be viewed in magnification from the side of the outer-casing on a small screen by means of lens-imaging technology. The other side lacking electrical connections has connections for the refrigerant pipes from the tiny compressor/refrigerator pump installed inside the case. inside the “cube-chip is a miniature power-plant which converts a proportion of the waste-heat into electricity which it uses to help power itself; thus saving energy from any external source once the chip has reached its thermal-equilibrium.

In this way a matrix of 64 x 64 2.8GHz hyperthreaded processor-cores can be utilised with a single motherboard; giving a total of 4096 cores in a single unit no bigger than 6cm cubed…And then someone manages to stick two of them together, then four…