End of Moore's Law and future processor stagnation *spawn
- Thread starter bbot
- Start date
hows that ?
Is molecular or quantum computing ready to take traditional silicon-based computing yet? The increase in computing power will stop l after limit of silicon-based computing is reached.
.
Avenues currently under pursuit includes graphene transistors, which I believe can be made smaller than when using silicon. Considering how thin graphene is, maybe stacked transistors will finally be viable. That could lead to the proverbial quantum leap in computing density.
They can probably gain a lot from developing better chip designs and improving memory throughput and latency. Even if computer chips couldn't shrink they could probably double the performance from the packaging and chip design technology catching up with the current process node. I have a feeling that chip design could get a lot better but much of the improvement of late has relied on simply making smaller/faster features.
Graphene does not have a bandgap. Which means it is very hard to turn it off.
MoS2 is also single sheet but has a bandgap.
I think it bears reminding people: Moore's law never talked about shrinking or anything like it. It specifically dealt with transistor count doubling every two years. You don't technically have to shrink a design to double the transistor count (see also: dual core, quad core, octal core chips) but it certainly makes it more cost efficient.
In practice, in some cases you really have to. For instance, NVIDIA needs a shrink to double the transistor count of GK110 in its replacement, because it's already reticle-limited (or just about).
Of course, there could also be process changes that increase this reticle limit, but that's not anticipated at this point, as far as I know.
Yes, it is about shrinking. You can build a bigger die and double your transistors, but that die will be more expensive.
Moore's law says that for a fixed cost of building a die, each shrink delivers about 2x transistors and each shrink comes around every year or so (when he wrote the paper). Dennard later on (I think) laid down the scaling laws for the other relevant design parameters.
Fixed cost is the whole point of moore's law. You don't get a law of your own for saying that you can buy 2 for twice the price.
Feel free to quote Moore where he talked about shrinking in his "law". Also feel free to point out the fixed cost portion of it too, while you're there.
The statement was doubling of transistors, that's it. The rest of the baggage that you describe is not part of it at all, in fact the cost (for a very long time) was absolutely NOT static, nor was the die size, and yet a lot of it happened on the same lithography size.
This is exactly why I felt like a reminder was in order, as a lot of things get attributed to Moore's law that really aren't part of it at all. Some folks say a doubling of performance, some folks equate it to equal capability at half cost, some others equate it to half power. Some of those are things that have come to fruition over the years, but they aren't part of Moore's law.
The statement was doubling of transistors, that's it. The rest of the baggage that you describe is not part of it at all, in fact the cost (for a very long time) was absolutely NOT static, nor was the die size, and yet a lot of it happened on the same lithography size.
This is exactly why I felt like a reminder was in order, as a lot of things get attributed to Moore's law that really aren't part of it at all. Some folks say a doubling of performance, some folks equate it to equal capability at half cost, some others equate it to half power. Some of those are things that have come to fruition over the years, but they aren't part of Moore's law.
Chip sizes have practical limits too and have more or less stopped growing, so Moore's Law has to be about shrinking now. It doesn't really matter what the situation was several years ago - if you talk about doubling transistors today you talk about shrinking. It's not the same as mistaking Moore's Law to mean doubling performance or halving costs.
This could change with stacking, but I'm not sure if Moore's Law still applies there.
This could change with stacking, but I'm not sure if Moore's Law still applies there.
To be perfectly clear here, Moore's law hasn't changed. What has changed is our implementation of the law, That might be simply larger chips (500mm^2 chips were simply not possible in his day, duh obvious statement is obvious right?) or stacking them three dimensionally, or shrinking them.
Our ability to meet that law is unrelated to the law itself, which is verbatim: transistors per device will double approximately every two years. Gee, how'd we do that? Maybe with bigger chips, smaller transistors, stacking transistors, maybe even a substrate that isn't silicon.
I don't think there is a line. It was an observation he made 50 years ago, i don't think it needs to operate in the bounds of a mathematical proof.
So it's whatever people want to claim to best make a trend line fit?
I agree though, it's just an observation.. and to some extent maybe had a cultural impact on Intel's manufacturing decisions (and by extension a secondary impact on everyone else). But some act like it's a law of nature. The semiconductor industry won't scramble to keep up with it by any means necessary, and new technologies won't happen to emerge just in time to allow for the next doubling in transistor count (and no more).
I probably have to read the original verbage of the "law", but what was it even tracking? An average of all chips, or just the largest product? And just from Intel? The fulfillment of the law managed to transition from improvement primarily due to increasing die size to improvement primarily due to shrinking transistors (with both happening in between) but I wonder if that's really much more than a coincidence. There's been definite market drive to allow shrinking transistor size at a regular schedule. I'm sure there have been a lot of planned advancements to allow for larger die sizes too but you don't really hear about them.
According to Wikipedia, the wording of the original observation was that for devices with the lowest cost per transistor, the complexity--with transistor counts now being the apparent measuring stick--would double every two years.
This is for integrated circuits, but possibly not necessarily the same piece of silicon since Moore made an additional note that he expected it could be done with a single wafer.
Moore's observation applied to the scaling of transistor counts for devices with the best cost per transistor.
What may be interesting to see, aside from whether this cost/density curve can be readily maintained, is whether with the diversification of design parameters and physical and economic challenges that devices that Moore's observation applied to are as relevant as they once were.
Even if Moore's Law does exist by some tortured definition for some set of devices, devices with the necessary attributes that allow a chip to meet the transistor/cost metric may not have the performance, power, or utility to be considered viable products for an increasing swath of products.
Worrying about Moore's law may become academic before it becomes physically and economically impossible.
This is for integrated circuits, but possibly not necessarily the same piece of silicon since Moore made an additional note that he expected it could be done with a single wafer.
Moore's observation applied to the scaling of transistor counts for devices with the best cost per transistor.
What may be interesting to see, aside from whether this cost/density curve can be readily maintained, is whether with the diversification of design parameters and physical and economic challenges that devices that Moore's observation applied to are as relevant as they once were.
Even if Moore's Law does exist by some tortured definition for some set of devices, devices with the necessary attributes that allow a chip to meet the transistor/cost metric may not have the performance, power, or utility to be considered viable products for an increasing swath of products.
Worrying about Moore's law may become academic before it becomes physically and economically impossible.
Similar threads
