very time I
buy a new smartphone, I expect to get more capability with faster speed, yet I
still expect all of this at the same price or lower. Why do I expect this to be
possible? As transistors get smaller, cheaper and faster, this should translate
to better technology each time you buy a new device. I guess that is why no one
wants to keep a phone until it dies anymore, and why we are used to getting
more for less each time we purchase a new phone.
relationship between transistor size and value has been somewhat constant over
the past decades and is commonly known as Moore’s law. While we may not be totally aware of how this law works, we have
come to expect the benefits that it has delivered. Roughly stated, approximately
every two years the density of computer transistors on computer chips doubles. This
means that computer chips take up less space and do more work at faster speeds
and at a lower cost for everything that uses them.
of transistor size have been used to describe the advancement in chip manufacturing
capability. Most recently these processes have been called 20 or 22 nanometers (nm), with these numbers describing the size of chip features – wires
or transistors – which can be easily measured. This provides a measure of
speed, cost and performance.
That is all changing according to Chenming Hu, the co-inventor of the FinFET state-of-the-art transistor. “Nobody knows anymore what 16 nm
means or what 14 nm means,” said Hu, when describing the latest FinFET chips
slated to begin shipping in 2014.
In a recent
article Rachel Courtland*
points out that chip makers are expressing the relative merit of new
technologies with “Node Numbers.”Design and process improvements drive each
improvement in performance while delivering better cost performance with no
relation to the size of the transistors or features used to make up the complex
computer chips. The relationship between the size of transistors and
improvements in density and performance is now related to other process
improvements. Since 2007 the size of transistor gates remained somewhat static
and in 2013 the pitch of wires on chips seems to have stalled. So chip manufacturers
now use these Node Numbers to indicate a new generation of technology.
track of Moore’s law will be more difficult in the future without the obvious
points of reference we used in the past such as transistor size. Perhaps new
metrics will become apparent. For example, effective chip density may be a new
metric to measure future node improvements. The IT industry is somewhat driven
by the technology progression predicted by Moore’s law, and it is an important
consideration in IT investment strategies.
Data center network
infrastructures are another example of this discussion. New technologies like software-defined networks are creating new capabilities by
enhancing traditional internet protocol data networks. Traditional data center IP
networks must now focus on other metrics – density, scalability, reliability
and agility – to support evolving data center network applications. The old
metrics used to measure network capability, such as 10G and 40G Ethernet, still
exist as a baseline, but new metrics are needed to describe new data center innovations.
SDN and other new data center architectures will drive better network efficiency.
As always, the founding physical layer must evolve to support these innovations.
Whether you are
buying a new smartphone or building a new data center, we all expect to see
better results for less money as time goes by. Innovation will continue to
provide cost improvement and help sustain the Moore effect for IT systems. Investing
in the right partnerships that foster innovation will serve you well as more
focus is placed on optimizing traditional technologies. Some of the traditional
drivers, like transistor size, might not provide the same pace of innovation
they have in the past. I believe Moore’s law will continue to hold true. I am
already planning my next system purchase based on what I expect the cost and performance
to be in two years.
How are you planning your future systems purchase?
*IEEE Spectrum, Nov 2013, pg. 24 - 27