Copper in chips and diamonds in drives lead to a shiny future
Fundamental technology rarely jumps so far and fast that it affects the computers you will buy in the next few months. But several such advances have occurred recently and, together, they could drive radical performance gains. No. 1: IBM Corp. plans to make silicon chips with copper microcircuits. Chip advances generally require either making components smaller so signals travel shorter distances or timing multiple operations within a single clock tick. The IBM plan is more
Fundamental technology rarely jumps so far and fast that it affects the computers you
will buy in the next few months. But several such advances have occurred recently and,
together, they could drive radical performance gains.
No. 1: IBM Corp. plans to make silicon chips with copper microcircuits.
Chip advances generally require either making components smaller so signals travel
shorter distances or timing multiple operations within a single clock tick. The IBM plan
is more fundamental and could bump up performance.
Until now, chip manufacturers have used aluminum, which is a better conductor per pound
but not better for a given cross-section. Electrons travel faster through copper than
aluminum, so less copper is needed for circuits. Also, components can nestle closer
together. These two changes could raise chip performance by 40 percent, according to IBM.
Copper chips run cooler, use less power and might cost less. IBM will begin producing
the copper chips next spring and doesn't intend to license the technology to other makers.
No. 2: The Energy Department's Lawrence Livermore Laboratory has come up with a way to
harden hard drives.
One bottleneck to packing more data onto hard drives has been the magnetic read-write
head's distance from the disk surface. The head can read only large magnetic domains. The
domains could be closer if the head could distinguish them.
When a rapidly spinning disk gets too close to the head, it touches and crashes. Even a
particle of cigarette smoke is larger than that gap, so bringing the head and disk any
closer together gets tricky.
What the Livermore lab, IBM and University of California at Berkeley collaborators have
done is deposit thin coats of synthetic diamond on the disk and the read-write head using
a cathodic arc. Crashes cause less damage because the disk and head are so much harder.
IBM's drives will store up to 2.6G of data per square inch without the diamond coating.
Now, 5G per square inch has been achieved in the lab with magnetic domains packed only 25
nanometers (25 billionths of a meter) apart.
To read such domains, the disk and head must be so close that they almost touch all the
time. Current coats, between 12 nanometers and 15 nanometers thick, keep the disk and head
too far apart. The diamond coating can be much thinner and harder, promising data
densities as great as 10G per square inch at little increase in cost.
Greater data density means more data on small drives, faster access and higher data
transfer rates.
No. 3: Ever heard of RDRAM? It's a memory advance that's been in the pipeline so long
that some computers already use it.
Dynamic RAM must receive periodic refresher signals that tie up processor cycles and
slow access time. Static RAM demands a steady power supply to retain data. It's about
three times faster than DRAM but uses a lot more power. Video RAM, which does simultaneous
read-write operations, is about twice as fast as DRAM and suitable for fast video refresh.
RDRAM (RAMbus DRAM) is so fast--10 times faster than DRAM--that it must be installed in
a specially designed motherboard or accessory board. But it eliminates the need for cache
memory.
As every recent PC buyer knows, that alone should lower system costs substantially.
RDRAM might even take data bandwidth into the gigabyte/sec range, although it only
functions at about 500 megabytes/sec today.
RDRAM is proprietary to Rambus Inc. of Mountain View, Calif., but it's cheap and very
fast, which almost guarantees acceptance. It's just now finding its way into mainstream
computers, starting with low-end graphics cards. Intel Corp. expects to supply RDRAM
boards for PCs before 2000.
Digital video disks, 3-D graphics and ultra-high resolution displays will demand
3-gigabyte/sec bandwidth by 2005. RDRAM is the only serious contender capable of reaching
that level. SLDRAM (SyncLink Consortium DRAM), a possible alternative to RDRAM, hasn't
shown as much progress. If you want to see RDRAM in action, look at the incredible
performance of the Nintendo 64 game machine.
I predict we'll see big advances from these three computer fundamentals, above and
beyond the so-called Moore's Law doubling of performance every 18 to 24 months, as stated
by Gordon Moore back in 1965.
John McCormick, a free-lance writer and computer consultant, has been working with
computers since the early 1960s. E-mail him at powerusr@penn.com.