Sep 292009

You may know of Moore’s law, which predicts that electronic transistor density will double every 2 years. This prediction has been accurate for decades, and is often taken as a law of nature. The results: cheaper, faster computers every year are a staple of modern technology and life. We expect constant improvement will continue indefinitely. But how long can this continue, and what should we expect to see years down the road?

This rate of progress – doubling every two years – is an exponential trend, which translates into very rapid improvements. However, in the real world exponential growth is always quickly reigned in by physical limitations. We do not yet have control of particles smaller than electrons, so there are limitations to the sizes of our transistors. We are already approaching hardware built from single layers of atoms which harness electrons one-by-one. Once we’re there, either computers stop improving or we find a new hardware paradigm (quantum computing?).

Remember that processor clock speeds used to increase year after year: we saw 500MHz, then 1GHz, soon 3GHz, a few at 4GHz, suddenly nothing new. Manufacturers and consumers calculated the value of a computer simply by the processor frequency, which was often incorrect and ultimately a failure. A limitation imposed by physical properties (temperature and voltage) resulted in no further possible frequency improvements. Suddenly, the industry had to rethink how the future would play out – Intel originally thought they would be able to take the Pentium architecture to 10GHz!

Although we haven’t hit a similar limit in transistor density, and analysts expect another decade of advancement, we should be prepared to hit this brick wall. There will be a point where we can no longer add more computational capacity to what we currently think of as a processor. And at the same time, we will not be able to pack any more data into a piece of computer memory.

Plot of future computer memory prices

As shown above, memory prices fit an exponential curve. As transistor densities double every couple years, memory prices fall. Although memory capacities follow a similar curve, market factors determine the actual sizes available (most vendors manufacture 1GB, 2GB, 4GB and 8GB DIMMs at the time of writing). Desktop computers use lower-cost, smaller memory than those plotted here. But high-end servers used in the fastest supercomputers in the world pack as much memory as possible into a single computer. That makes this plot something of a worst-case scenario, as home users will almost certainly be able to purchase lower-priced memory than shown here.

The trend is shown through 2030, but it is not likely current technology will hold out that long. Analysts are currently predicting a breakdown somewhere after 2020, although these breakdowns have been predicted before and some major advances are likely (multi-layer silicon?).

Here are my predictions (see you in 10 years…):

  • 2010: 16GB DIMMs @ $1,200
  • 2013: 32GB DIMMs @ $950
  • 2015: 64GB DIMMs @ $1,100
  • 2017: 128GB DIMMs @ $1,200
  • 2020: 256GB DIMMs @ $1,000
  • Maybe 2030: 4096GB (4TB!) DIMMs

  One Response to “The Future of Computer Memory”

Comments (1)
  1. We’re still on track! 32GB DIMMs have been shipping for a while and cost about $1000.

 Leave a Reply

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>

(to be displayed above your comment)

(for moderation; your e-mail address will not be made public)