Moore's Law states that the number of transistors installed on dense integrated circuits double every 18 months. Gordon Moore made this observation in 1965, and technology has been progressing along these lines ever since then. Engineers and even software developers have found that the law is precise enough to accurately predict the constant march of advancements in the field of semiconductors.
That steady march is starting to slip however. Practical considerations mean that semiconductor chips are unlikely to see any real gains in transistor numbers after the year 2020. R&D crews have been hard at work to come up with entirely new paradigms. This means that completely different computer architectures are about to come into play. Several of these architectures are fast approaching maturity. Others won't come on the market for years to come.
Most semiconductors are very literally hardwired to perform in a certain expected manner. Discrete logic gates can only move in the direction that their solder points allow them to. Field-programmable gate array (FPGA) technology allows engineers to put together integrated circuits that designers can configure after they ship. This technology has actually been around for a very long time. Steve Casselman applied for a patent for an FPGA chip that implemented 600,000 re-programmable gates back in 1992, and since that time these chips have been used in several fields. Industrial, automotive and telecommunications buyers have driven the market for traditional FPGA chips.
New developments have made it so that this architecture can serve as a replacement for existing discrete chips. Most modern FPGA chips feature the ability to reconfigure themselves at runtime. This has lead to developers designing machines that reconfigure themselves to suit whatever task is at hand. An FPGA-based machine would therefore be able to run a legacy mode in order to provide compatibility for older software. Operators could then turn off legacy mode to run modern operating systems and take full advantage of the resulting speed gains. Most importantly they could include both analog and digital components.
Stacking silicon wafers on top of one another is a new way to make advanced integrated circuits. They're then connected through a series of vertical through-silicon via wires that cause them to act as a single microchip. Some motherboard systems are unable to distinguish between traditional microchips and those made through 3D stacking techniques.
These certainly allow more functionality to fit in a smaller space, which extends Moore's Law through a whole new generation of smaller devices. Keeping signals inside of a single chip as opposed to spreading it through several devices can also reduce energy consumption by several orders of magnitude while increasing bandwidth at the same time. This also means that circuits assembled in this fashion generate less heat. This means that they can be run at a much higher clock speed without risking thermal failure. It opens up new opportunities for people who want to overclock their machines too.
The most mature newer architecture relies on multi-chip module (MCM) design, but this kind of technology probably wouldn't sit well with Gordon Moore. Engineers adhering to this design install various existing microchips into a single package in order to make hybrid integrated circuits. While this doesn't make for a very compact package by comparison, it's still small enough to fit into game consoles and mobile devices. Desktop computer manufacturers are currently looking at linking 10-20 individual existing microchips to form radically new architectures that won't require major leaps in engineering. Most of these designs allow engineers to buy chips right off the shelf.
Quantum computing was a fantasy until recently. Real quantum chips turned out to be nothing like those developed in the minds of science-fiction authors. Current designs are centered on wire-bonded superconductors that form logic elements inside of a mounted module. These chips count quantum bits rather than electrical states, which means that they're far more flexible than standard digital microchips. It also means that they're among the smallest designs possible.
Optical chips use the same technology that powers videodiscs, but it does so on a much smaller scale. These chips work via a nonlinear system, but they're actually not as different from existing microprocessors as one might imagine. Light is made up of electromagnetic waves, so the basic idea isn't much different. Transistors used with these chipsets are based around nonlinear optical effects, but they still adhere to the same logic gate paradigms seen in a majority of regular digital computers.
