BL602/BL604 RISC-V WiFi and Bluetooth 5.0 SoC will sell at ESP8266 price point | Hacker News Reader

BL602/BL604 RISC-V WiFi and Bluetooth 5.0 SoC will sell at ESP8266 price point

cnx-software.com

214 pointszdw58 comments

We've seen ESP8266 and ESP32 succeed tremendously in the maker space. We've yet to see widespread adoption in commercial and professional applications - that's coming, but I am surprised why IP for bluetooth/wifi is so expensive outside of Chinese chip makers? ST, NXP, Microchip, Renesas, etc. have really expensive SoCs - $15/chip and over. These giants seem to be sleeping. What's preventing them from competing with the Chinese?

Also, we have horrendous price gouging for memory. A mere 128k to 256k bump in memory spec on a microcontroller will send you back a $1.20 in chip price.

Is there some CCP-gov subsidy going on to capture the market? Espressif's ESP32 is such an extraordinary chip with 500kb+ memory and yet, it is an order of magnitude cheaper than anything close from western suppliers. What gives?

The trick is that ESP* chips implement a lot of the stack in firmware. This saves a lot of IP in dedicated hardware that allows for price reduction, but has it’s drawbacks, mostly in terms of energy consumption.

There’s also a good amount of problems involved in having a non-auditable (non certified) piece of firmware running in a coprocessor that handles communications, and being closed source and Chinese doesn’t help.

The difference is generally that the maker space doesn’t need to be production ready.

By the way, Murata has modules on the ~6$ range (depending on what features you want)

Not only the radio is software, cores themselves, as I understand, are not hard macro, and I find it more surprising that such very generic synthesized chip is not an even bigger power hog.

Just throwing the entire design at the synthesizer these days seems to be pretty close to a collection of hand laid out hard macros. The synthesizer being a lot more aggressive about optimizing out signals that leave the macro (and downstream uses of those signals, so those global optimizations that are hard for a human to see) seems to make up for it's deficiencies with perfect local layout.

Wait what? You are telling me the ESP's are actually FPGA's with a preconfigured core?

I probably have no idea what I'm talking about because I don't work in this space, but in a previous life, our company was looking at Nordic for chips, and they had some pretty solid offerings for bluetooth chips. IIRC it was around $5 per chip but I may be totally wrong. It was one of these I think, and it's been about 3 years since I looked into this https://www.nordicsemi.com/Products/Low-power-short-range-wi...

Nordic has some great parts and their tools and stack are first-rate. No WiFi though, last I looked anyhow.

Right. We particularly cared about Bluetooth 5 + USB, NFC and cryptography chip was a bonus.

Yeah, it's interesting considering that supposedly engineers are not significantly cheaper in Chinese tech hubs compared to the West.

The Western chipmakers are being disrupted with chips that are good enough, not realizing that cheap price creates a new quality, enabling devices not possible before.

Sure, the firmware and SDK had some of bugs and missing features to iron out, but they came a long way since the release and now are solid enough to put devices like Sonoff inside your electrical boxes and walls.

We had issues with Simcom modem chips as well as Telit. The newer Simcom chips are better and quite competitive, pre Sim800 it was riddled with bugs. If then Telit was an option, now it isn't because it's considerably more expensive and the IoT market is very competitive. 3G chips were the worst pricewise because they had to include royalties to Qualcomm, that's why most chipmakers pushed for 4G.

The ESP family is absolutely amazing in terms of price and features, not so much efficeincy wise. It would have been great if RTOS support was better. The ESP-IDF is based on FreeRTOS, NuttX has some preliminary support, RiotOS doesn't do FPU, BT, crypto acceleration and uses one core only. NodeMCU is maker/hobby garde, not an OS and so is Arduino. To be fair, the whole ESP chip family is rather hobby grade but if it works and you're not power constrained or you use it as a secondary MCU and put it to sleep, it should be fine.

> NuttX has some preliminary support, RiotOS doesn't do FPU, BT, crypto acceleration and uses one core only.

As always, patches welcome ;)

Those RTOS projects don't have nearly as much developer resources as there are Chips, so there are always still low hanging fruits to be picked.

Until very recently ESP chips couldn't compete with Western chips you mentioned in various scenarios, and in some they're still out of the question. This is a gradual process though and Microchip & co. will try to keep the prices of their wiresless chips high as long as they can.

> What's preventing them from competing with the Chinese?

Greed

> Is there some CCP-gov subsidy going on to capture the market?

Every chipmaker in China has a 10 year 100% income tax subsidy. Some of the earliest subsidy recipients are now passing that period, so they invented new schemes which are to the effect the same except for "go to technology park A, or factory park B."

Given this is an income tax, companies don't benefit much themselves if they are working on low margins, but it helps to lure in more companies who got themselves into stockmarket play.

As it is China, another factor is that company owner who is a sole shareholder/board head/CEO would've been taxed to oblivion if he were to simply give himself a giant salary to extract profits (60%-70%.)

When it all began, dividends were completely tax free, and there were no capital gains tax. Now they aren't, but the policy somehow continues to drive on the same rails.

> Also, we have horrendous price gouging for memory. A mere 128k to 256k bump in memory spec on a microcontroller will send you back a $1.20 in chip price.

To be fair, the esp* chips don't have memory mapped on-chip flash but load the code from external SPI flash to RAM at startup.

While that solution is cheaper (and also faster due to executing from RAM) it adds a small delay at startup powering the large RAM uses more energy.

Esp8285 has 1MB integrated flash memory (2MB on one variant).

https://www.espressif.com/sites/default/files/documentation/...

it's build into the package, but it's still a separate die connected via SPI.

Yes, from a power standpoint that matters. From a BOM cost and PCB layout standpoint, it doesn't.

Espressif shipped 100M units by Jan 2, 2018. I’m not sure what they’ve shipped in the last 2.75 years, but I’d wager it’s at least another 100M.

That’s not just to makers for sure.

https://www.espressif.com/en/news/Espressif_Achieves_the_Hun...

You see them in a lot of home automation gear, like wifi power outlets

> These giants seem to be sleeping. What's preventing them from competing with the Chinese?

Realtek has been selling the rtl8710 for a long time, however it seems like it never really caught on (there have been some experimentations with Rust on it in the hobbyist world, but that's all I know).

https://www.cnx-software.com/2016/07/28/an-alternative-to-es...

I was under the impression that tuya is moving from esp8266 to rtl8710 for their IoT ecosystem (which in turn includes many third parties, who buy their firmware + smartphone apps + cloud infrastructure from them). This primarily affects those who wish to flash tasmota on e.g. their wifi-controllable sockets.

Another board with interesting features that never caught up was the WiFiMCU, based on the EMW3165 SOC module by Mxchip.

http://www.wifimcu.com/ (Lua programmable just as the NodeMCU)

https://en.mxchip.com/modulelist

Walking through my local hardware store (in NZ) there's a bunch of cheap wifi controlled wall outlets - every one I've looked at has an esp8266 in it - maybe it's Trump's bogus trade war that's keeping them out of the US

I think it will be interesting to find out which product used ESP8266 based ESP-01 and -12E modules for their original intended purposes before it was “found”. They were meant to be low performance UART/SPI/I2C/SDIO... compatible “modem” to cheaply add Wi-Fi to some other product, probably some tablets or IoT home hub or appliance device.

They were already mass produced and available as factory leak/direct sale/replacement parts at dimes apiece, then were picked up by the guy who ported Arduino Core to it when he was looking for anything small and cheap that has Wi-Fi and are reprogrammable.

It could have been some Sierra Wireless UMTS or Intel Wi-Fi bgn or CSR Bluetooth, but the easiest he could tap onto was an 8266. I remember there was a guy showing demo doing what today one would do with M5Stack done on a classic miniPCI card in Maker Faire Tokyo like a decade ago, except notably the logo on the guy’s badge and laser etching on the chip kind of looked similar so that kind of future could have been possible but never materialized. Similarly I still have couple “Zero Channel RAID” cards somewhere that I dreamed of turning them into what Raspberry Pi is today. So a lot of these opportunities exist/ed but are/were neglected for the lack of interest to get over a chasm. Perhaps modern day equivalents to them are DRAM cached SSDs, some of them smell like penguins.

Why do chips like these not have 5 GHz Wifi?

Living in a city, the 2.4 GHz band is crowded, and a lot of devices require you to play with your router settings to set them up [1] e.g., some Mill heaters must be set up on a 2.4 GHz network using a channel below 10 [2]

[1] https://millheat.zendesk.com/hc/en-us/articles/213305665-My-...

[2] https://millheat.zendesk.com/hc/en-us/articles/213310805-WiF...

Because everything else is on the 2.4GHz band and so adding 2.4GHz Wifi doesn't require any new HF circuitry.

That wouldn't apply to Espressif since they started with WiFi, right?

You can make 2.4ghz radio on more or less the same process as the logic.

With 5ghz, silicon cmos is still possible, but the power consumption will be very bad. The need for a specialty RF process is a big limiter.

What I mean is that you will have to go for 2 pieces of silicon, (one for radio on a specialty process, and another for regular CMOS logic) instead of one.

Even desktop wifi chips are still made in 2.4ghz only variants to save some cost in the lowest end, and for the chip priced in cents, it is completely not an option at the moment.

A lot of the problem is often not the band is crowded, but that often one too many send out DeAuth packages. This is fixed in WPA3.

The setup issues aside, these devices need very little data, have small antennas and are often in hard to reach places.

2.4ghz with its better coverage is actually more appropriate.

There's probably also the issue that there is no _one_ 5GHz band. Suddenly you have to deal with DFS radar detection, and different regulatory compliance domains. It now matters which country you're selling to, or which firmware they're using.

Probably because the target price point is at or below $1.

You can get a RTL8720DN that has Dual-Band WiFi for $4

https://www.cnx-software.com/2020/01/17/realtek-rtl8720dn-du...

Finally a promising RISC-V ESP32/ESP8266 alternative. My question now is if the ADC is any good. Not so on the ESP's.

12-bit ADC vs ESP8266's 10-bit (and similar to ESP32's 12-bit). Interesting thing is that BL602 has a 10-bit DAC!

No ESP32 contender, half the RAM and no I2S peripherals means audio applications are out. Too bad.

Half the RAM might be OK given that it seems to follow the 'execute out of a flash cache' rather than 'loading the image fully into RAM' strategy the ESP chips use. I could def be misunderstanding it though.

And I2S being gone is very unfortunate, but I like chips with upstreamed LLVM backends enough that a really cheap CPLD as a SPI/I2S bridge wouldn't be the worst thing in the world for my designs I use for at home projects.

It supports XIP QSPI Flash (https://en.wikipedia.org/wiki/Execute_in_place). I wonder, wouldn't the loading of instructions be limited by QSPI speeds? (Much less than CPU clock speed)

There's nearly always a cache sitting between the flash controller and the CPU when an XIP flash scheme is used, normally integrated in the flash controller itself.

BL602 ordered from AliExpress (BL604 is not available). Looking forward to long range connection between BL602 and RPi 4.0 (that also use Bluetooth 5.0)

It appears that the Raspberry Pi 4 does not support Coded PHY for extended range or any optional Bluetooth 5 features. Many Bluetooth 5 chips out now do not support those optional features, so you need to confirm the physical layers available to see if a device is suitable for your project.

How do you expect to program it? Have you found any English documentation anywhere?

Is it illegal to import the HiSilicon Hi3861 (which seems to be a more powerful chip) into the US?

I can see modules with that chip are just a few dollars on alibaba:

https://www.alibaba.com/product-detail/Hi3861-Wifi-Chip-Hikv...

The Bluetooth and WiFi stacks look proprietary, that's unfortunate. Anyone knows if there's any documentation or source out there about interfacing with the WiFi/BT peripheral?

Can we abuse this as a SDR ?

Why not use a chip intended for that?

https://www.microchip.com/wwwproducts/en/AT86RF215IQ

It's less than 4€ and you can build and SDR with it

https://tinysdr.org/

https://github.com/uw-x/tinysdr

It will be great to have another cheap microcontroller in the space with much more advanced radios. Is there any info on power draw? I didn’t see that from the initial specs.

It's right there on the page but it looks like he blocks copy and paste which is ridiculous: 0.5 uA (deep sleep), Wifi Rx: 18 mA. The first link (https://www.cnx-software.com/2020/10/12/hi3861-based-hispark...) has more information.

I'm puzzled about his Rust remark; if it's an RV32 then Rust already supports nostd on riscv32i-unknown-none-elf, riscv32imac-unknown-none-elf, and riscv32imc-unknown-none-elf.

If those figures are right, that would put this new chip as dramatically better than esp32, right?

Yes, the esp8266 and esp32 are power hogs and run quite warm, which is why you don't see them in battery powered gadgets very often. It'll be interesting to do real world measurements when this chip is available.

I think i saw the rust thread on twitter. It was about Ferrous system supporting this board with their course and cli tools.

Probably also with a peripheral access crate (type-safe register definitions) and and HAL integration.

So, there's language support for it, but it'll still be necessary to have driver support, ideally a whole support library in Rust.

Most BLE or WiFi enabled microcontrollers have a binary blob for driving connectivity, or deeply C based code that can be a challenge to integrate idiomatically in Rust.

I don't speak Chinese, but this linked page seems to have some information on the power draw: https://www.ednchina.com/news/2020SONGSHANHU-3.html . I don't know if that is just the WiFi component, or the whole SoC. You can also see two other kind of SoCs they contrast their BL602 offering against.

The biggest problem of ESP32 is power consumption on BLE fein my point of view. Would really like to see a competitor where ble runs with an ULP or so.

What kind of tooling / libraries are used with these? Eg IP or 6Lowpan stack.

Reply on news.ycombinator.com