Chapter 4: The language had owners ― the x86 and ARM empires

In the previous chapter, we found that the ISA (instruction set) is "an invisible shared language, and contract, that meshes hardware and software." So then — whose is that shared language? In fact, for a long time it was a "possession" held by particular companies. In this chapter, we talk about the two empires that have divided the world. It becomes a bit of a story about power and money.

4.1 Two "languages" that divide the world

Until now, most of the world's chips have spoken, roughly, one of two ISAs (shared languages).

Open a PC, it's x86; hold a phone, it's ARM. Without knowing it, we use one of these two empires' languages every day.

4.2 It costs money to use the "language"

Here is the core. Since these shared languages have "owners," using them involves cost or a license. The way differs by empire.

x86 is, in effect, a language where "only insiders may speak it." In practice, only Intel and AMD can make it. Even if a new company thinks "I want to make x86 chips too," it cannot, as a rule, join that circle. It is a very closed world.

ARM, on the other hand, is somewhat more open: the business is "pay a fee, and we'll lend you the design." Many companies pay ARM a license fee and use that shared language (and design) to make their own chips. The door is wider than x86's. But the one holding the "source of the language" is ARM, and the structure of paying a toll at the gate is unchanged.

To put it another way, a "toll bridge." To cross to the other side of the river, you have to take that bridge. The bridge's owner can take a toll from those who cross. As long as the bridge is convenient and everyone crosses there, money comes to the owner without pause. To hold the "language" that is the ISA is like owning a bridge everyone must cross.

4.3 The one who stands in the middle holds the value

Let's widen the view a little. In the world, there is a common structure: whoever secures "the place everyone must pass through" gets a large share. The entrance to a market, the payment system, the platform … The position of standing in the middle, between the makers and users who sweat in the open, and taking a toll, is strong. The ISA was one of these.

The person who designs the chip, the person who writes the software, the person who provides a service on top — all of them have no choice but to ride on the same "language." The one who holds that language, even without making the final product itself, can draw up value, little by little, from the whole ecosystem. This is a power of "structure," separate from whether the technology is good or bad. And structural power often lasts longer than technology.

4.4 Why could a small, unknown company hold sway? ― ARM's story

Here is the most interesting true story in Chapter 4. ARM, the world's most widely used language, was not a giant from the start. Its beginning was a small company in Cambridge, England, where a very small number of engineers took on the large companies. When it became independent around 1990 as a joint venture with Apple and others, it had just over a dozen engineers. It could easily have vanished. So why could it come to hold sway over the world? What decided it was, more than technology, a choice of "structure."

• It happened to be "low power." As a by-product of a simple design with a pared-down instruction set, its power consumption was low. A plain strength at the time turned, later, into a decisive weapon in the age of the mobile phone.
• It decided "not to make chips." It stopped earning from its own products, lent the design (the language) to many manufacturers, and took a small fee per shipment. = It moved to the side that builds the bridge and takes the toll (4.2).
• It made rivals into allies by "letting them use it." For many companies, CPU design is not a point of differentiation. Borrowing is cheaper than in-house development, and frees up capacity for their strong areas. The more adoption grew, the more the benefit of being common (richer tools and software) grew, like a snowball.
• Being neutral, anyone could ride it. A position belonging to no particular semiconductor company. ARM is often called the "Switzerland" of the semiconductor industry (which echoes neatly with RISC-V's move to Switzerland in Chapter 5).
• It hit big with mobile phones. Low power told in phones, it took the standard seat, and the snowball became decisive.

※ "The one who doesn't make them reaches the widest." About 35 billion ARM chips are now shipped a year, iPhones among them. Yet ARM itself ships not a single chip. What it receives is a fee of roughly 1 to 2% of a chip's value — thin, but because it's a bridge everyone must cross, dust piles into a mountain. This is the extreme of 4.3, "the one in the middle holds the value."

4.5 Why doesn't this empire crumble easily?

"If it's closed / tolled, someone should just make a freer language" — you'd think so. But it doesn't crumble easily. The reason is the "pain of moving house" we saw in Chapter 3.

On top of x86 and ARM, decades' worth of accumulated software is stacked. The OS, the apps, the business systems. To switch to a new language, that vast asset must be rebuilt. And a chip of a new language, whose software isn't ready, is wanted by no one at first. Then a "no chips so no software / no software so no chip sales" chicken-and-egg deadlock arises. This strong "lock-in" has long protected the two empires.

4.6 The Apple case ― famous for going its own way, yet the language is ARM (and it has switched before)

"Aren't the chips in iPhone and Mac proprietary?" is a common question. This is half right and half not. What Apple makes itself is the "guts of the chip (the design)"; the language (ISA) it speaks is ARM. It borrows ARM's language, then builds, from scratch, a "body" that speaks that language — which is how it achieves "fast yet low-power." Even Apple, famous for going its own way, is a resident of the ARM empire when it comes to the language itself (this proves 4.4's "the lender is strong," from the reverse side).

More interesting still is that the Mac has switched its "inner language" many times through history. In Chapter 3 we wrote "moving an ISA is hard," yet Apple actually did it, more than once.

   The Mac's "inner language (ISA)" changed four times
 ──────────────────────────────────────────────
   1984   Motorola 68000 family (68k)
   1994   PowerPC (the Apple-IBM-Motorola alliance)
   2006   Intel (x86)              <- "the Intel era" is here
   2020   Apple silicon (ARM-based) <- now, here
 ──────────────────────────────────────────────
   With each switch, it prepared a "bridge (translation)"
   to run old-language software on the new language (e.g. Rosetta).

Two things follow. One, an ISA is not forever; the victor changes with the era — even on the Mac, the now-dominant Intel (x86) was a resident of just one era. Two, switching is possible, but it takes the large apparatus of a bridge. This is living evidence of Chapter 3's "pain of moving house."

4.7 And there, one question was born

By here, the stage is set. The shared language has an owner, a toll is charged, and lock-in keeps it from crumbling easily — this was the "given" until quite recently.

But someone posed this bold question.

   What if --
 ───────────────────────────────────────────────────────────────
   that "shared language"
       belonged to no one,
           needed no toll,
               and anyone in the world could use it freely?
 ───────────────────────────────────────────────────────────────
   If the bridge toll became zero. How would the world change?

It may sound like an outlandish ideal. But this question did not end as a fantasy. From a university lab, such a "language" actually came into being. In the next chapter, Chapter 5, we enter the story of this corner's lead — RISC-V. But first, there's one important caution to pin down: "free" and "open" are not the same thing.