6 ms·

The main reason semiconductors scaled the way the did, was due to economies of scale. The more refined the process, the more R&D money had to be put behind the

• BeeOnRope 8y ago
  Moore's law was very different because the underlying process was being scaled. CPUs just process information, and the inherent limit on the amount of "stuff" needed to perform say an addition has an incredibly small physical limit [1] which existing chips don't approach. So through scaling the feature size of chips, CPU manufacturers were able to take the same amount of silicon, the same size wafer, and get "double" the computational power (roughly speaking) out of it. So with about the same amount of "assembly line" work you can churn out something that is twice as fast in 18 months. After a decade its 100 times as fast, but your "assembly line" looks about the same. That's not "efficiencies of scale" - that's fundamentally making something much better due improved physics, with the same amount of work. It particular, it would apply even to "small" producers. Now don't get me wrong, CPU manufacture was also subject to traditional efficiencies of scale: the biggest fabs got bigger, and a few large players squeezed out the rest and were able to sell more and average our their R&D costs over more sales, but that effect is small compared to the million-fold improvement in the underlying physical design. Batteries have no such scaling. The power stored is basically related to number of lithium ions and the capacity of the anode to accommodate them all. There are some small efficiencies: you can increase efficiency from 80% to 90%, but never to 2000%. You can make materials thinner or cheaper. You can change form factors to use materials more efficiently. You can standardize on battery sizes to make more use of a single production line. You can secure long term lithium contracts and open more mines. These are all the traditional "efficiency of scale" things and they all hit a wall pretty quickly. You could probably easily sustain a 15% reduction for a while longer, but certainly not 40 years like Moore's law. > The same principally applies to almost anything. It doesn't, just look around. What else has decreased in cost by a factor of a million over the past few decades? Cars have increased in popularity by many-fold since the 40s , and after an initial period of traditional "efficiencies of scale", costs have remained relatively fixed. Look at any random food product that suddenly increases a surge in popularity, perhaps reaching a 10x sales multiplier: final cost and production costs don't drop 10x. If all of a sudden we start eating 2 avocados every meal they aren't going to start costing 10 cents. > In fact, for typical physical products the saying is that double the production will halve the cost. I can imagine this rule is true... up to a point! That's the "traditional efficiencies of scale" at work: it's usually an S-curve [2]. If you want some custom widget, you are probably going to have to pay $100 for the first one, and $1 each or whatever for your run of 100. When you order one million, maybe it drops to 1 cent. When you order a billion, they don't cost 0.001 cents though. You don't have to take my word for it though: just look at any two big companies, but where one is bigger than the other, and look at their costs of production. Let's say Coke sells 5x as much as Pepsi: does Pepsi cost 5x as much to produce? Not all, they are virtually identical. Many models of vehicles sell 10x or 100x of some unpopular rivals, but the production cost is about the same. --- [1] https://en.wikipedia.org/wiki/Landauer%27s_principle https://en.wikipedia.org/wiki/Landauer%27s_principle [2] Approximately, at least - although it depends heavily on the product. For example, the initial part of the curve might not be very flat for some things (e.g., with large fixed one-off costs) - but they almost all share the "rightmost" flat part of the S-curve.
  • heisenbergs 8y ago
    While the scalability behind silicon was perhaps easier and faster, there are similar effects at work for batteries. Li-ion batteries are typically spooled in layers [1] (see https://en.wikipedia.org/wiki/Lithium-ion_battery#/media/File:Lithium-Ion_Cell_cylindric.JPG https://en.wikipedia.org/wiki/Lithium-ion_battery#/media/Fil...). The thinner the layers, the higher the capacity as you have a smaller distance between positive and negative layers. From my discussion with battery chemists, this is what is primarily driving the annual ~15% reduction, economies of scale also obviously playing a role. Unsure what the theoretical limit of battery size layers is, and how far off we are from that, but the thinner we get those layers, the higher capacity the batteries per weight and also presumably price. [1] https://en.wikipedia.org/wiki/Lithium-ion_battery#Shapes https://en.wikipedia.org/wiki/Lithium-ion_battery#Shapes
    • BeeOnRope 8y ago
      > While the scalability behind silicon was perhaps easier and faster, there are similar effects at work for batteries ... The thinner the layers, the higher the capacity as you have a smaller distance between positive and negative layers. I'm arguing it's not at all all similar. The capacity of a lithium ion battery is fundamentally limited by the amount cathode material it contains, along with sufficient electrolyte to support it, just like any other battery. Making other materials thinner, allows you to stuff in more of this stuff, and other changes in the arrangement may make the process more efficient, but up to a limit. There is hard cap to usefulness of all of these processes, at the "theoretical efficiency" and in principle softer caps before that point long before you approach the theoretical caps. If you have a chance, ask the battery chemists you know what the best-case storage is for a particular chemistry and cathode material, compared to what is available today. I don't think it is more than 10x aware and is probably much less. That's very different than CPU scaling where you started out many billions of times away from the physical limits of the computational capacity of the material, and many trillions away from the theoretical physical limits of computation. So no, thinner stuff isn't going to sustain a 15% annual increase in efficiency [1]. In fact, I don't think think it will even sustain a single 15% increase from today until the end of time. Of course, there are many other vectors along which battery efficiency can increase when the denominator is "cost", even if the storage/size doesn't increase much. You can increase manufacturing efficiency. You can introduce new form factors. You can tweak the chemistry. You can carefully match the required discharge characteristics to the application. Outside of the battery itself, you can improve charging and discharging algorithms, you can use finer-grained control over smaller groups of batteries, you can improve thermal management. However, these are just in the range of normal industrial optimizations that apply to any product. You can replace "batteries" with "lettuce" in the above and come with a similar list. --- [1] To be clear, there is no 15% increase in battery efficiency per year, when measured by volume, weight or other physical characteristics: that stat must involve "price". Panasonic, the best and biggest Li-on player out there, has barely budged in efficiency on their headline battery, the 18650.