How is it possible that the ESP32 is so cheap?

[greengnu]

Even if the whole wireless functionality would be left out we’d still have a dual core 240mhz 32bit cpu that can run complex applications for a dollar (soc). I couldn’t find any other cpu in the same performance class that comes even close in terms of price performance.

How is this pricing possible?

Just saying. If I were the chinese government and wanted to spy on the world through an armada of small IoT devices linked into every private wifi around the globe....

[ESP_Angus]

Folks, discussing conspiracy theories is entirely off-topic for this forum and if necessary posts in this thread will be moderated.

Even if the whole wireless functionality would be left out we’d still have a dual core 240mhz 32bit cpu that can run complex applications for a dollar (soc). I couldn’t find any other cpu in the same performance class that comes even close in terms of price performance.

How is this pricing possible?

The short answer is that it's cheap to manufacture. In particular the RF engineers have done a bunch of very clever things on the Wi-Fi side. You will also notice that in a lot of ways ESP32's design is not like other common microcontrollers. This is generally not by accident, it's to keep the overall cost down.

Just saying. If I were the chinese government and wanted to spy on the world through an armada of small IoT devices linked into every private wifi around the globe....

This conspiracy theory comes up every now and again, and as a core Espressif engineer I've tried to figure how it would technically work. And I have no idea. Would all ESP32s ping a server once a day to check if they should switch into "backdoor mode"? And what would "backdoor mode" do exactly, given every chip has different peripherals attached to it in different ways? And noone would notice this additional unauthorised behaviour on their WiFi, despite millions of chips in daily use?

Keep in mind that all network code in ESP32 above the Wi-Fi MAC layer is open source, the WPA2 integration is open source, and the build system and toolchain are all open source. So the binary Wi-Fi libraries somehow have to implement an entire "shadow" firmware implementation that could take over the main firmware, no matter what that main firmware happens to do, and without any ability to analyze the real firmware source. Then they'd all have to phone home with no interruption to normal device behaviour and nothing unusual that shows up on a TCP packet dump.

Engineering a chip that is cheap to manufacture seems positively simple by comparison.

[greengnu]

as a core Espressif engineer I've tried to figure how it would technically work. And I have no idea. Would all ESP32s ping a server once a day to check if they should switch into "backdoor mode"? And what would "backdoor mode" do exactly, given every chip has different peripherals attached to it in different ways?

Well, this https://www.bloomberg.com/news/features ... -companies chip was able to do it somehow too, and it's a fraction of the size of an esp32 and was embedded into a much more expensive, much lower volume product. The penetration of the esp32 is likely going to be many orders of magnitude higher than this and basically all of the devices that contain it will already be network connected.


Don't get me wrong - I'm absolutely amazed by the ESP32. I love this little device a lot! I'm just getting suspicious when something seems too good to be true...

[WiFive]

If a 200mm processed silicon wafer costs $1000 and esp32 has a die area of 8mm2 then the cost per chip is somewhere around 30c so it is possible. You might have to sell 5MM to break even on NRE but espressif has sold 100MM chips (all products). Another reason why it is so cheap for everyone is because there are no traditional (Western) pricing tiers.

Also I think that Bloomberg article has been debunked many times over but it gets clicks which is all that matters today.

[greengnu]

If a 200mm processed silicon wafer costs $1000 and esp32 has a die area of 8mm2 then the cost per chip is somewhere around 30c so it is possible.

By that logic a 28-core intel xeon platinum should cost 20$ instead of 13'000$.

How do you pay for all the engineering, marketing, legal cost etc. etc. that are required to get something like the esp32 out?

[WiFive]

Well Xeon Platinum is a much more expensive silicon process with lower yield so it probably does cost over $500 in just silicon.

[ESP_Angus]

How do you pay for all the engineering, marketing, legal cost etc. etc. that are required to get something like the esp32 out?

Investors.

I'm locking this thread because speculative conspiracy theories are off topic for this forum.