Cost of the Future

I’m concerned that every hardware device being released by startups right now costs between $100 and $200. Certainly, used or older models of devices can be found for slightly less, but at that point they are generally no longer supported by the company and/or no longer have full functionality because current applications, firmware updates or necessary cloud accounts no longer exist.

I wonder what kind of qualitative changes need to take place in the space of hardware device engineering that would allow for a (relatively) smart device to be built at an order of magnitude (or two) less in cost than devices today.

By example, this light switch is available for $0.69 at full retail price. The only “smart”-ish light switch I could find quickly was the Belkin WeMo Light Switch which retails for $50.00 at full retail price.

I find this example particularly disturbing for several reasons:

  • The only onboard communication the device has is WiFi. There’s no Bluetooth, ZigBee / 802.15, FM, or anything else - just one radio and no hard-wired network.
  • There’s no display
  • There is zero need for any kind of computation whatsoever outside of the network stack (zero vacuum tubes needed)
  • It’s a fucking light switch.

Now, I don’t mean to pick on Belkin here in particular - in fact, I think a more modern take on this would probably just include a couple more radio options or small display and double the price.

I think there’s an “engineering” culture that has kind of fallen apart. Without a doubt, the technologies and attitudes we’ve used in building out the Web and it’s enormous ecosystem of API-driven widgets have brought us a long way. The difficulty I see is that if we’re essentially looking at REST protocols over WiFi as the level we need to interact with a light switch, then either:

  • We double-down on this approach

  • WiFi modules need to get a lot cheaper (see our comparison)

  • Consumer WiFi routers need to be able to handle hundreds of connected clients without melting

  • We need to be able to run a bare-bones REST server without involving Linux or a real OS at all. Preferably, handle this at the firmware level, if at all.

    • Said hardware needs to be a lot cheaper. This can be accomplished by abandoning clock speed, as well as size, caching, and most optimizations. In this case, the processor couldn’t cost more than $0.10, and the WiFi would have to max out at around $0.50.
  • We admit that baking fourteen levels of abstraction* on top of a powerhouse chip and 100+ metre radio is irresponsible at best.

It’s of course easy for me to favour option (2), but more difficult for me to propose what the ideal solution would look like. I think the real answers here lie not necessarily in new technology (except where absolutely necessary for cost-cutting, like perhaps 802.15), but in finding older technology still in plentiful supply that can find gainful re-application in ultra-low-cost electronics such as light switches.

Bring on the $5 connected light switch!


*There’s a real cost to layering too much abstraction - especially when we’re talking about hardware devices. As ugly and multi-layered as modern web protocols like WebSockets happen to be, we’re able to generally bring the computing power to bear on the problem, along with incredibly comprehensive debuggers and sysops on the server side to keep the magic alive.

When it comes to the Internet of Things, added levels of software abstraction directly correlate to energy and processing requirements, as well as all kinds of radio and network overhead… and those things directly impact cost in a (usually) exponential way.

3 Likes

In the short term, I think we’re going to see more and more wifi SoCs - like the ESP8266 and the TI CC3200. Running your application on the same chip as your comms and rf output reduces cost, size, complexity, and power usage.

I think the only sane communication solutions at the moment are MQTT and CoAP. Both are low ceremony, low overhead protocols that were made for this sort of embedding.

The ESP8266 is currently available on $3 modules. I’m not sure what the chip costs in bulk, but it can’t be much more than a dollar or two. Seems to me the way forward is all Wifi SoCs until we figure out a way to get off of TCP/IP. Maybe that’s MQTT-SN? Maybe we stop needing embedded TCP/IP when our phones can talk 802.15.4/.5?

It’s nice to see things like the ESP8266 out there - but it does seem like the overhead of 802.11 / WiFi (technically, energy, computation, and license cost-wise) puts a high lower-limit on the size and cost of the device. Even $2 is a lot if you’re looking for a dumb link to put on many small things.

I’m very excited about 802.15 in general, but I don’t think TCP/IP is going to go away – it’s just too insanely successful (and useful!). The ZigBee IP spec which is going to support IPv6 over ZigBee seems like a very good solution if implemented natively.

The reason I see TCP/IP as so crucial is that it’s what made the Internet really work in the first place, and really allowed networks to connect to each other on a first-class level. If we go with non-TCP (or even non-IP) solutions for personal networks, there will always be a sort of awkward impedence mismatch that some widget is going to have to do translation over.

My only caution here is that when people think TCP/IP, they tend to have a significant software stack in mind. The last thing we need is an operating system running inside my light switch.

My latest project to pick on: a guitar cable over WiFi costing over $200.

What annoys me is that market forces haven’t yet caught up to the point where this kind of thing is just ridiculous. From a components perspective, this really shouldn’t cost more than $10, even given today’s chipsets and protocols… but it takes effort to drop an order of magnitude off the cost, and apparently we’re not demanding that effort.

Well, it comes down to what do you want? What should we have in a consumer level home network? Do you want to have generalised networking infrastructure like wifi for everything where you can guarantee some basic interoperability, or a bunch of (quite likely clashing) specific radio networks, each with gateways that all need setting up and maintaining?

No one RF comms solution is going to cover all kinds of use cases for al sorts of applications perfectly (one of the reasons zigbee is still doing so poorly after all the effort and time chucked into it… it was designed from the ground up to be data transport for a home automation network, and still, it only is really compelling for use cases where you have one of the profiles that just do what you want. If you want to do something the people behind zigbee alliance didn’t think of during their network design, you are pretty much out of luck) And to setup and commission a new non-trivial network is still a serious undertaking for someone who knows what they are doing - to adjust the network as a user, say to add or remove or change around the purpose of an endpoint, it’s pretty painful.

If you want to use something ubiquitous like wifi, you get a certain level of flexibility that cheaper and simpler and lower power networks will never provide. but that costs you at a BOM level and an energy use level. As already mentioned, those new CC3200 wifi SOCs from TI look awesome, I bought 2 dev boards to play with recently because they are going to be able to do a lot of very cool stuff very easily - but you won’t be seeing just a chip like that for the cost of a conventional wired light switch for years… and they still use real energy every time they wake up, find a wifi network, get authenticated, find their target, open a session with their target, send their 1 bit of actual information that is their data payload, receive the ACK, log off the network, and then go to sleep.

Also I think it’s a bit unfair to look at something that’s basically a pure mechanical part (lightswitch is a couple of injection moulded bits of plastic and some stamped bits of metal, with a couple of machined terminals.) all put together automatically using a couple of machines you buy once, and feed basic commodity materials into indefinitely. Compare to the complexity of an RF based system that will need plastic, switch, PCB, batteries, the radio, etc etc etc… plus all the specific design work, legislative compliance, automated testing, supply chain issues, future part EOLing, etc etc etc… it’s a completely different game.

Then we come down to the real killer here - how useful is a wireless light switch in the first place?

The only advantage I can see you get with an RF lightswitch over a wired one, is you can install and move them about as you see fit… but if light switches are installed when buildings are constructed, well, the cost of wiring them in isn’t that much. and then you have something that works forever.

The disadvantage of the RF lightswitch is batteries will always need changing (unless you run power to them… hmmm…) and the fact that RF signals will always be subject to interference, blocking, and channel fading depending on how the RF environment they are in changes. And that’s really not cool for something you depend on as much as a light switch.

Even at the same price as a conventional switch, without shifting the entire paradigm of what a light switch is and does, should we ever expect wireless light switches would sell even a remotely similar quantity to the simple reliable mechanical ones? Even with the entire paradigm shifted, would you expect wireless switches to outsell ones with a simple RS485 or ethernet hardwired connection that also allows continual powering?

Maybe the real trick here with wireless IoT gizmos is to create new products with compelling value, so the $100-200-ish cost is seen as being worth it, rather than mindlessly taking something that already works perfectly well, and making a wireless/IoT/smartphone attached version for the sake of it?

3 Likes

These are all fantastic points.

Very true.

I don’t think wireless is inherently better for these applications, but rather I think we’re just going to see this class of device skip over wired in favour of wireless much like the developing world with telephones. Especially as the next wave of these devices will still be purchased by consumers prior post-construction, it’s unlikely they want to deal with wiring.

If I were able to engineer the entire electrical/networking system, I agree that a wired solution would be better in almost every respect.

So, this is also a very good point. However, my point is really targeted at things that aren’t special - devices that are super ubiquitous. And my feeling is that to meet that vision for the future - where really everything connectable is connected - we’ve got to do something serious about cost.

The slightly more unfortunate bit is that I think nearly all existing IoT gizmos have totally failed your test: wireless kitchen scale at $150? wireless guitar cable at $200? light switch (and now also wall socket) at $50? These devices are reselling for between 10x and 100x the price of their unconnected counterparts.

I feel like IoT gizmos are still targeting the sort of airline magazine diamond-encrusted-toaster market, rather than more mainstream markets where consumers don’t experience violent buyer’s remorse.

I think the idea of having all the home automation appliances connected over Wi-Fi is a bit ridiculous. There are a number of problems in my mind:

  • Cost - the main theme of this thread
  • Reliability - have you ever waved your phone around trying to get it to connected to a weak access point?
  • Usability - do I really have to give my Wi-Fi password to each light switch?
  • Security - can I really trust these devices to be secure from attack?

Hopefully Power Line Communications / Broadband over Power Line can solve most of those problems. With that, any mains connected device can have a reliable connection with IPs handed out by DHCP which helps with my reliability and usability complaints. Unfortunately you still would need to buy a BPL adapter to go between your router and mains, but I think you’d only need one per house.

I have to wonder whether powerline networking is feasible vs this. I’ve yet to see a good cheap powerline solution (or even a bad cheap one). Once you’re at ‘needs a hub,’ it always seems that there’s some RF tech that looks more attractive. Whether that’s crazy range or just cheap with great functionality, I don’t see a good reason to pick powerline.

That said:

  • I wonder if powerline requires intentional radiator certification; if not, that could be a nice upside.

  • If there were a SoC (or even a module) that’d do online SMPS and powerline networking, I’d be pretty excited about it.

The fucking light switch could at least go down to a cost of $15. Have you heard of the q88 tablet? Well, it’s not really one tablet, but it’s an OK tablet for $15-25 (Hua Qiang Bei prices, but I think you could do similarly well online). It’s a start, anyway…

1 Like

Can anyone say how much these CC3200 chip is going to cost on bulk, i checked out at few websites like Digi key and element 14 but the quote of around $11 per unit doesn’t make sense to me. Any device you make with these device may be any home automation device such as light switch or smart plug are not going to be sold at a price of $11 and these chips are not so suitable for ip camera’s, i found in ti’s website with a quote of around $7|1000 units what this means $7 per each unit or $7 per 1000 units?