As things currently stand, the flash controller is internal to the M1 SoC. This results in significant cost and power savings, as well as some rather impressive performance.

Using an external SSD format like M.2 means the added monetary and power costs of 1) an external flash controller, 2) the mechanical components to interface with it, and 3) multiple PCIe lanes which can never be fully powered off.

> the flash controller is internal to the M1 SoC

Okay, it was part of the T2 chip before. And they were able to put replaceable SSDs into the Mac Pro. Those only contained the NAND chips and were proprietary, but still.

The problem I have with soldered SSDs is that flash memory has a limited lifespan in terms of writes. Once you're out of writes, then what? This is aggravated by the fact that M1 can't boot even from external media if the internal SSD isn't working.

It's worth noting that Apple's SSDs tend to be on the high side of the industry for write endurance. I think their calculus is that no reasonable user would exhaust the lifespan before the rest of the computer is well and truly obsolete.

(This was briefly not the case with that Big Sur SSD thrashing bug, but that's been fixed now)

It's actually split. The NVMe controller is in the M1 SoC, but the flash modules contain their own dumber controller and connect via PCIe in the back end (using a custom protocol, not NVMe).

Apple could and have put their SSDs in modules - that's how the T2 Mac Pro does it. They're still proprietary though, of course, due to the split design.

I don't. If on chip, yes. But in the same place with the same speed interface it should not matter once the wires are connected.

You can't take on-package RAM and stick it on a separate module through a connector (what connector? You need a 512-bit interface for the M1 Max...) without causing a problem or massively increasing your power consumption, for the same reason you can't string together 30 meters of USB3 extension cables and expect it to work. This is why USB 3.2 chipsets are little bare chips and run at 10Gbps, while 10Gb Ethernet chipsets need a big heatsink. More distance means a higher power consumption for the interface.

Interfaces are designed for specific characteristics like maximum capacitance and signal loss, and going beyond the design parameters means you need to change something. In this case, it would involve significantly increasing the power consumption of the interface.

Even if the SSDs were modular, you still wouldn't be able to do offline data recovery on them, because they're encrypted with system-tied keys. So this is tangential to them being replaceable or not.

>I can understand the soldred RAM on M1 — yes, speed of light and other laws of physics get in the way

so you do know the speed of signal in copper? ~20cm per 1ns. 5-7cm of extra track length to a slot would not hurt anything.

It absolutely does. It adds capacitance, which means you need stronger drivers and higher voltage levels, which means you end up with a quadratic power increase to maintain the same performance.

You cannot push a 512-bit wide memory bus like on the M1 Max through a connector and longer PCB traces without massively increasing the power consumption of the interface. Remember, those laptops have a memory bus equivalent to 8-channel traditional DDRx RAM. You'd need 8 SO-DIMM slots just to get the bandwidth, at a huge increase in power. LPDDR RAM isn't even available in modules for this reason; it's designed for low power, and using connectors goes against that design goal.