The Old Joel on Software Forum: Part 3 (of 5) - Death of Moore's Law greatly exaggerated

Death of Moore's Law greatly exaggerated

"Intel, the world's biggest maker of computer chips, has achieved an average feature size of just 65 nanometres for its next generation chip. The memory cells produced are half the size of the most advanced manufacturing technology in use today."

http://www.newscientist.com/news/news.jsp?id=ns99994424

I'm always amused for the past 30 years pundits have predicted that next will definitely spell the end of Moore's Law.

Matthew Lock
Tuesday, November 25, 2003

Not enough coffee this morning. I meant:

"I'm always amused how for the past 30 years pundits have been predicting that next year will definitely spell the end of Moore's Law."

Matthew Lock
Tuesday, November 25, 2003

I wish we had something like Moore's law for disk I/O.

Mike
Tuesday, November 25, 2003

We have inverted Moore's Law for disks. Instead of 200% every 18 months, it's 18% every 200 months. :-p

Brad Wilson (dotnetguy.techieswithcats.com)
Tuesday, November 25, 2003

You mean disk transfer rate? Well your memory is already bigger than hard drives were 10-15 years ago, so in some sense you could say the transfer rate has vastly increased, while the capacity remained the same. Though of course memory is volatile, but that doesn't really figure into the argument, since you can of course load all of memory from disk or some persistent storage once.

Roose
Tuesday, November 25, 2003

Moore's Law is on a collision course Stein's Law: "things that can't go on forever, don't".

With mathematical certainty we can know that Moore's Law has a limit. The only question is when that limit will be reached.

Alyosha`
Tuesday, November 25, 2003

If Moore's Law has a limit, it's not based on mathematics (since you can double a number forever), it's based on physics. And that limit would be based on physics as we understand them today, assuming we maintain the same methodology of building CPU circuits.
We've got a ways to go before we even have to start thinking about quantum CPU's.

Einstein-RosenRosen
Tuesday, November 25, 2003

Also Moore's law can double if the technique improves, even if we hit a brick wall in physics, such as 3d chips etc.

Matthew Lock
Wednesday, November 26, 2003

hell...harddrives are sufficiently small now that its not unreasonable to get at least one more application of moores law just by doubling the physical size of them (or build a 2-in-1 type system).

FullNameRequired
Wednesday, November 26, 2003

What we need is some kind of technology to allow us to build an array of inexpensive disks.

Tim H
Wednesday, November 26, 2003

Clearly, the reduction in integrated circuit feature size is reaching a limit.

From an economic viewpoint, the "end of Moore's law" isn't a brick wall, a circuit density below which integrated circuit production is easy and above which it is impossible. Instead, what is happening is that the costs of creating the integrated circuits is increasing. At high densities, I've heard of mask sets costing $1 million, because small feature sizes make producing masks so expensive.

The implication is that high volume commodity parts will be produced with very high densities, but that lower-volume parts and ASICs may not be.

An interesting result is that FPGAs are becoming more attractive as replacements for other hardware. FPGAs are high-volume commodity parts, and reaping the benefits of Moore's law. I'm amazed by the density of Xilinx parts, and the low cost (once you back off of the top-of-the-line parts.)

The hardware design languages (VHDL and Verilog) frequently used for FPGA design are essentially programming languages enhanced to describe hardware structures. For instrumentation designs, I'm seeing computer/microcontroller/FPGA architectures in which the software/hardware boundary is very mushy: the application software, the microcontroller firmware, and the FPGA layout are all written in high-level programming languages.

Dan Brown
Wednesday, November 26, 2003

[Editor's note: Acronyms should be defined the first time they're used unless they're widely known. This is most strongly recommended when the acronym is the central point of your writing]

Philo.
Wednesday, November 26, 2003

ASICs and FPGAs are pretty well known acryonyms in the chip design world.

But to handicap for you software freaks: ASICs (application specific integrated circuits) are custom chips built from the gate level up, whereas FPGAs (field programmable gate arrays) are chips full of basic logic blocks (ANDs, ORs, XORs, adders, flip-flops, etc), and all the designer has to do is wire them together -- not as optimized as an ASIC, but much cheaper to make.

Alyosha`
Wednesday, November 26, 2003

The other part of Moore's law was that the investment required also doubled every eighteen months.

How many of you can remember the time when you would get cloned Pentiums - my first computer in 1996 had one.

They rapidly became uneconomic to produce.

Stephen Jones
Thursday, November 27, 2003