Microchip's 50th anniversary

Today marks the 50th anniversary of Jack Kilby of Texas Instruments demonstrating the first microchip.

I like TI. My first proper calculator was from Texas Instruments. In the mid 1990s they also invented the micro-mirror, which is now used in a lot of video projectors, and I believe will evolve into the basis of most future active contact lens displays, raster scanning tiny laser beams onto our retinas.

Computing today would still be a toy for big science and the military were it not for the microchip. As the basis of almost all of our IT, as Gartner's Jim Tully observed "integrated circuits are so woven into our lives that it would be hard to imagine a world without them". And certainly we have a great deal to be grateful for, with the quality of our lives far higher than would have been the case without chips.

Although I have a lot of respect for Gartner, and I fully applaud the huge impact the chip has had on our lives to date, and am happy to agree with their short term prediction that the number of integrated circuits produced annually will rise to 330 billion by 2012, it is always a good idea for futurologists to explore the future without prejudice, and most empires crumble eventually. It is a bit early to say that chips have had their chips, but they will not be around in today's form for ever.

So what of the future? Moore's Law will continue for a good while yet, but miniaturisation of circuit components will hit the stops eventually due to quantum effects. Using multiple layers as well as multi-cores will take chips gradually away from the 2D limits of today. As the number of cores multiplies, in due course, it will make good engineering sense to sever the direct links between the circuits and to suspend them in gel, where cooling will be easier and free-space optical interconnects can replace the complex on-chip wiring that takes up so much space
and creates so much of the manufacturing difficulty. It is then that the integrated circuit itself comes into the final phase of its empire. Once circuits are no longer directly hard wired, self organisation can come into play. Starting with a soup of a family of generic circuit components, self organisation can configure them into sophisticated ad-hoc circuits from the ground up, and reassemble them as the task permits. That means that we will no longer need to have a fixed physical chip layout constraining the behaviour of our devices. Hardware level evolution can be utilised along with software, to experiment with different circuits and processes to achieve a variety of tasks. This approach would be excellent for devices such as robotics, where they can develop appropriate control and sensory interactions to get the best out of their physical capability, learning how to use 'limbs' and so on. As insights come in from neuroscience, direct implementation of these into hardware initialisation, followed by evolutionary tweaking, could quickly emulate a lot of the sensory and cognitive processes used by nature. Results can feed positively back into neuroscience and AI, accelerating the design loop even further.

I believe this will be the route for most strong AI development. We don't know how consciousness works, and conventional research is slow. Utilising a strong positive feedback loop of neuroscience, nanotech, AI, self organisation, evolution and gel computing, the promise is much greater. Gel computing allows the much greater flexibility of configuring circuits that can be complex mixtures of sensors, actuators, memory, digital and analog processing, and adaptive neurons. We will accomplish intelligent machines, conscious machines, a long time before we fully understand their working principles.

I have long since estimated that we will see the first conscious machines with human level intelligence some time between 2015 and 2020. Most other IT researchers think this is ridiculously optimistic, but I stand by my estimate and see no reason to move it yet. Not much has happened so far towards that goal except the weak signals on research direction, but that is always the case right up to the last few years of development when fast exponential or even super-exponential development is involved. The first 10% of the work may have taken a million years. The last 90% will take a few months - in a few years time.