A new technology could make building computers easier and more efficient

MAKING a computer out of bricks sounds an odd idea. But that is one of the latest proposals to come out of IBM's Almaden Research Centre in San Jose, California. If the centre's Collective Intelligent Bricks are a success, they could turn building data-storage centres--and ultimately, perhaps, supercomputers--into a game of LEGO.

Collective Intelligent Bricks are cubes with edges about 20cm long. Inside a brick there are 12 disk drives, each of which can store 80 gigabytes of data. The bricks also contain a processor to run the disks and a chip that sends signals to the connectors that join a brick to its neighbours.

Those connectors are novel, too. Each face of a brick has 16 metallic pads on it. These pads match up with those on neighbouring bricks to form small capacitors. An electrical signal on one side of a capacitor induces a similar signal on the other side, so the capacitors allow the bricks to talk to one another without the need for cumbersome cable links.

At the moment, this capacitive coupling, as it is called, is able to transmit data at a rate of 3 gigabits a second. The researchers at Almaden hope to push that up to 10 gigabits a second soon--a rate comparable with the best optical-fibre connections. Jai Menon, who is in overall charge of the project, expects to have a working demonstration of a 3x3x3 array of bricks before the end of the year. This would have a capacity of some 25 terabytes--the equivalent of all the information in America's Library of Congress.

Besides being compact, this design has one of the properties that engineers value most--redundancy. If one of the bricks goes wrong, the network of connections between the others can be re-routed to isolate it. And because the whole system can store so much information, back-up copies of everything in it can be made internally in more than one brick, meaning that such a failure destroys no data. Indeed, the system is reckoned by its designers to be sufficiently robust to tolerate multiple brick-failures over the years. This means that it will never need to be repaired--or at least that by the time it does need repairing it is likely to have been replaced by a new and improved product.

There are, inevitably, costs to such an arrangement. In particular, it is hard to cool since the bricks in the middle have no contact with the air. A system of pipes pumps water through it to carry the heat away, but despite the optimistic nickname "IceCube" that the researchers have given to their invention, this is undoubtedly a weak link in the design. If the pumps fail, so will the machine--probably irreparably. Nevertheless, Dr Menon is optimistic that the design will work not only for data storage, but eventually, once it includes bricks containing banks of processors as well as hard drives, for general-purpose computers as well.

These could be scaled up by adding more bricks. Expensive high-speed communications between processors and data-storage areas would be eliminated. And by stacking components so compactly, signal latency (the delay that results when data are moved from one part of a computer to another) would be greatly reduced, further enhancing performance. Bricklaying, in other words, will have become a high-tech profession.

Sep 11th 2003
From The Economist print edition