The ESP32 never ceases to impress me. They're ridiculously cheap and low power, and the fact that they're able to emulate a PC operating system just makes them even cooler.
ESP32 doesn’t impress me by the things you said. The most impressive thing about ESP32 is the ESP-IDF and the amount of effort Espressif has put in to make it accessible, developed in the public and amazing progress over the years. Compare that to TI or ST, they look like they’re stuck with management issues and not getting out of the rut. It boggles my mind that HALs are still developed in isolation and not open source. WTF are these companies thinking. Microchip PIC platform is another dumpster fire.
FWIW they have a RISC-V based ESP32-ish microcontroller now[1].
Espressif has excellent good low-level documentation and a ton of code on Github[2]. I haven't built a system up from code, but they seem as open as any microcontroller supplier without having to buy into an NDA.
What does everyone think about the ESP-IDF move to CMake? I personally thought their "legacy" Makefile setup was really nice and worked perfectly for me. What does CMake bring to the table that I haven't appreciated yet?
It would also be nice if their tooling was less Python-heavy, that's a common point of failure and slowdown when running commands.
Overall though I agree, the ESP-IDF is great and I wish more chip makers put a similar amount of effort into their tooling.
TI, ST and Microchip's primary customers (big OEMs) haven't cared about open source, Github available SDKs, etc. Things are changing but it's a slow ship to move.
I made the following comment once before on a thread about a business card with a processor running Linux:
In the late 1970s DigiKey was represented by a one-column ad in the back of Popular Electronics. A featured 7400-series chip with four NAND gates cost $0.25 . Inflation-adjusted, that twenty-five cents now buys enough processor and memory to run Linux.
The $99 Jetson Nano gets you a computer that is more capable in nearly every dimension than the 1986 Thinking Machines CM-2 super computer. The CM-2 had more RAM, though a lot of it was spread across its 65536 processing engines. The CM-2 is way prettier though.
This is exactly right. I really like the Cortex-M4 line as they have lots of peripherals, flash, and some have DDR controllers that will drive megabytes of RAM. On a cycle by cycle basis they are roughly 10x the performance of an Intel 80286, the chip in the PC/AT.
Indeed, that is why when people think the only way to code these chips is Assembly and C, while we were already using multiple languages on the PCs, shows how there is some kind of misunderstanding of their capabilities.
Heck, compiled xBase applications were hardly a speed daemon, yet they were fast enough for what businesses needed of them.
Yes, but without the pipeline and the complex instruction set!
That said, it will be interesting when one of these SOC vendors decides to make a 'south bridge' serial I/O bus (I'm thinking like four LVDS channels clocked at 100MHz)which then connects to a selection of peripheral modules. That would give customers a 2 chip solution that could have a lot more variety than trying to stock 100+ SKUs to meet various market needs.
Yes, although I'd go with the original (and simpler) Intel Hub Architecture concepts.
Interestingly enough you can do this on your own using the RISC-V architecture and FPGA's with high speed serial transceivers. So it isn't like you have to be a chip company to build something like it, it just means you're willing to spend way more to implement a RISC-V system than you technically "have" to :-).
If things are anything at all like they were when I was at Intel then it is an inability to handle risk. Wafer starts are expensive, masks are expensive, margins on this sort of processor are tight. Basically that adds up to nobody wanting to be "that guy" (and it is always a guy) who pushes through the idea that costs the company a few million dollars.
As a result, unless some customer both specifically asks for this, and commits to enough volume to insure that they will at least break even, they aren't going to risk it.
I also like the ESP32, but low power seems like a stretch... that thing runs very warm. I suppose if your benchmark is an i386 or a Raspberry Pi, then sure, it's low power, but compared to other microcontrollers with comparable wireless peripherals, it seems to consume more power than any of them.
It’s absurd, as a microcontroller. Like ~300mA max which is like 1W. Though only in short bursts if planned wisely, and also there are worse ones. I believe Nordic nRF does 2A bursts that are to be taken care of by a larger than usual bypass cap.
That doesn't sound right... all of the Nordic designs I've seen have been exceptionally low power. The internal LDO / DC-DC blocks for running directly off of a battery are sized for <100 mA. Even with bypass caps, 2A spikes seem like the wrong order of magnitude for a chip that can run off of a coin cell.
I have an nRF52840 design, and I've never seen it draw more than low ~tens of mA during BLE transmit spikes, even at the maximum transmit power setting.
It's one thing to say that there's more computing power in a coffee maker than would fit in a desk 40 years ago - but to actually demonstrate it is mind blowing.
Even a run of the mill smart phone can out compute IBM's Deep Blue's 12GFlop from 1997.
