Taiwan Semiconductor Manufacturing Firm (TSMC) simply held its 2024 North America Expertise Symposium, the place it crammed in attendees and traders on its know-how roadmap for the long run.
You could rightly ask: okay however what does this must do with me, a person of Apple merchandise? Effectively, TSMC is a chip fabrication firm, and sure probably the most superior one on the planet. They’ve been Apple’s companion for almost all of its chips main chips—the A collection for iPhones and iPads, the M collection for Macs, and extra. And Apple is usually first in line for brand spanking new manufacturing processes from TSMC, paying a premium to be the primary buyer to get to make use of 5nm or 3nm applied sciences, for instance.
Right here’s a abstract of the TSMC roadmap and what it would imply for future Apple silicon, and due to this fact future iPhones, iPads, Macs, and extra.
What’s a nanometer?
Earlier than we discuss TSMC’s future applied sciences, let’s have a fast reminder on what a “nanometer” is on this context. Technically, it’s one billionth of a meter. A human hair is between 50,000 and 100,000nm thick. Most micro organism are between 1,000 and 10,000 nanometers.
In silicon course of know-how, the “nanometer” measurement is how huge among the options of the chip are. Totally different corporations measure totally different options–it was once the size between the supply and drain elements of a field-effect transistor (FET), however nowadays totally different elements are measured by totally different corporations.
The A17 Professional chip was Apple’s first 3nm processor, adopted by the M3.
The A17 Professional chip was Apple’s first 3nm processor, adopted by the M3.
Apple
The A17 Professional chip was Apple’s first 3nm processor, adopted by the M3.
Apple
Apple
In different phrases, 5nm means some particular elements of the chip are solely 5 nanometers huge, however TSMC’s 5nm is just not the identical as Intel’s 5nm, is just not the identical as Samsung’s 5nm, and so forth. A smaller nanometer quantity means you’ll be able to match extra chip logic or cache or no matter in the identical quantity of area, which might result in extra highly effective chips, decrease energy consumption, smaller chips that match into smaller units, and so forth.
Consider it a bit like taking a look at a metropolis in Apple Maps–zooming out makes every thing smaller, placing extra buildings, streets, and land on the identical quantity of display screen. That’s what shifting to a smaller nanometer course of is like extra “metropolis” in the identical area.
There are a lot of different essential facets of a microprocessor, together with the way in which transistors are insulated, supplies used, and a lot extra, however the “nanometer” measurement has caught as a approach of differentiating one main manufacturing era from one other.
3nm progress
Apple was first with TSMC’s preliminary 3nm course of, which was known as N3. The corporate has now refined that with the N3E course of, which is what we predict Apple will use in its most superior merchandise this fall (A18 and M4). Whereas it might sound vital, the primary focus of N3E is to make the chips extra reasonably priced. There are some slight variations in density and efficiency but it surely’s not a significant generational change.
2nm coming subsequent 12 months
The following main change is the shift to 2nm, which is predicted to occur in 2025. Apple is as soon as once more anticipated to be the primary main (and presumably solely) buyer, so it’s potential that the A19 or different chips (perhaps an M5?) that ship in late 2025 will use this course of. All of it comes right down to TSMC’s means to work out the kinks in manufacturing and yields and such in time to supply tens of thousands and thousands of chips with it.
In comparison with the N3E course of, the N2 course of is predicted to cut back energy consumption by 25-30 % (for a chip of the identical complexity and frequency) or to enhance efficiency by 10-15 % on the similar energy consumption. Chip density (how a lot stuff matches right into a single space) is predicted to extend by 15 %.
An fascinating change to this era of chips, apart from merely being smaller/denser/sooner, is one thing TSMC calls “NanoFlex.” It would enable chip designers to make use of cells from totally different chip libraries all on the identical wafer. Normally, a chip designer has to make use of all blocks from a “low energy,” “excessive density,” or “excessive efficiency” library, relying on crucial wants of the chip. By letting designs use totally different elements from totally different libraries, chips can fine-tune totally different areas to their wants.
For instance, Apple might resolve it’s most essential to make the video and audio encoders and decoders a part of the chip as small as potential and lay out that a part of the chip utilizing the high-density design libraries whereas utilizing the energy-efficient libraries for the low-power CPU cores and the high-performance libraries for the efficiency CPU cores.
For the chips Apple produces, the limiting issue tends to be energy and thermal dissipation. So you’ll be able to in all probability count on chips made with the N2 course of to have extra “stuff” in them (cores, cache, larger and extra complicated video encoders, and so forth) to the tune of 15-20 %, with barely increased clock speeds and due to this fact efficiency, in comparison with chips from the 12 months earlier than. Nonetheless, the flexibility to optimize particular elements of the chip with instruments from totally different chip libraries has the potential to repay huge by way of increased “peak” efficiency or decrease idle energy.
The 12 months after N2 is launched, TSMC can have two enhanced variations of the method: N2P which is concentrated on prime efficiency, and N2X centered on decrease voltages and energy consumption. It’s unclear whether or not Apple will undertake a type of for the chips that are available 2026.
A16—we’re doing angstroms now?
The most important shift after 2nm (N2) is a course of TSMC calls A16 (no relation to the A16 Bionic). It’s a 1.6-nanometer course of however now that issues are getting so small, they’re form of getting off “nanometers” and switching to “angstroms.” An angstrom is a ten-billionth of a meter, or 10 instances smaller than a nanometer.
This one is just not coming till late 2026, nearly actually too late for Apple to make use of that 12 months. We’ll see chips made with the A16 course of from Apple in 2027, most probably.
TSMC gave some early estimates in comparison with the approaching N2P course of, the place A16 is predicted to enhance efficiency by 8-10 % on the similar voltage and complexity or scale back energy by 15-20 % on the similar frequency and transistor rely.
The large innovation within the A16 era will probably be bottom energy supply, one thing TSMC calls “Tremendous Energy Rail.” This runs a community for energy distribution on the again aspect of the silicon wafer, linked to the transistors by way of little tunnels by way of it. This improves density and doubtlessly reliability, as energy doesn’t must be routed round with all of the sign and clock distribution traces on the highest aspect of the chip. Different chip producers are pursuing related applied sciences (Intel’s PowerVia involves thoughts)—mainly totally different approaches to the identical concept.
TSMC could also be somewhat later than opponents like Intel with this type of know-how, because it has been pushed again somewhat. It was initially anticipated to debut within the N2P course of, and now will probably be first launched in A16 as an alternative.
Chips in Apple merchandise that use the A16 course of will be capable to have much more stuff (extra cores, larger caches) than the N2 course of whereas sustaining the identical energy profile.
Having chips with extra density or a greater energy profile sooner than others is one in all Apple’s huge benefits, however the actual magic comes from wonderful chip design and software program improvement that optimizes Apple’s software program particularly for the chips they produce.