Amazon Echo Flex Teardown: The Big Silicon

https://electronupdate.blogspot.com/2021/01/amazon-echo-flex-teardown-big-silicon.html

by parsecs • 6 years ago

119 57 6 years ago

electronupdate.blogspot.com

In ~1997 this level of compute power would have been enough to get on the TOPS500 list of the fastest computers our civilization is capable of creating.... now something that consumes only 54 sq mm of silicon that gets put into a $15.00 gadget.

... and it's incapable of doing anything useful without even bigger computers somewhere else.

Progress is so odd sometimes.

londons_explore• 6 years ago

I often wonder if there is anything else in society that could scale in a similar way to moores law.

Is there anything else we could use a lithographic-like process to make massive numbers of very cheaply, yet still find buyers for?

Floors of a modern skyscraper are nearly that - they're made with moulds that get reused again and again on every floor. But to get decent scaling, you really need to make instances in at least 2, and preferably 3 dimensions in one go.

We've already done it with things like the printing press that can churn out books very cheaply - but it turns out that we've now reached the limit of the numbers of books people want.

Humans also reproduce exponentially - with a sufficiently high birth rate, numbers of people can rapidly grow. One needs enough space to support that growth, and we are sadly lacking in that... Perhaps Mars will allow us to practice our fast-scaling of civilization?

TeMPOraL6• 6 years ago

Perhaps Mars will allow us to practice our fast-scaling of civilization?

For a few decades (assuming fixed rate), until the population doubles and we'll be overpopulating two planets. We need to either slow down on the exponent - not just population, but also markets - or figure out interstellar travel rather quickly.

bserge2• 6 years ago

I believe space would be the least of our worries in that case...

mistermann• 6 years ago

Ideas, emotions (love, hate, apathy, etc).

See: 1/6/2021 events at the US Capitol, the number of children that die annually due to malnutrition, etc.

baybal27• 6 years ago

Humans also reproduce exponentially - with a sufficiently high birth rate, numbers of people can rapidly grow.

With a sufficiently high birth rate, daycare, cheap real estate, schools with less than 100 pupils per class, free higher education, and solid, assured pensions, and state elderly care.

baybal27• 6 years ago

Is there anything else we could use a lithographic-like process to make massive numbers of very cheaply, yet still find buyers for?

Very few. The thing with litho is that materials inputs, and outputs are nearly the same, it's just the value coming defies conventional economics.

Thermodynamics limits all physical processes by entropy, but nothing limits the entropy itself.

mechwarrior• 6 years ago

Plastic molding of different types (injection, extrusion, blow molding, etc) enables massive quantities of physical objects to be produced at a staggeringly low cost and high rate of speed. This has of course enabled the production of many things that are beneficial to our lives. It has also increased the amount of waste that humans produce, to say the least.

bserge2• 6 years ago

Well it's not technically incapable, it's just that the makers can't profit from it.

gsich• 6 years ago

I'd call that regress.

1vuio0pswjnm7• 6 years ago

Sort the opposite effect with software. Something like "In ~1997, some of the largest desktop software applications would have been only a fraction of the size of today's smartphone apps."

ladyanita22• 6 years ago

I was thinking of the MS Office Android app. You can get it running on a $50 tablet. How powerful would it be vs let's say Office '97 on a Windows 95 PC purchased for around $2000 25 years ago?

remus• 6 years ago

I don't think it is that odd, it's just a case of costs, scale and who you're designing for.

A TOPS500 computer is gonna set you back a few million $, so you're only going to have one of them and it's economic to employ a team of well qualified people to write software and interface relatively directly with the hardware.

Now that same computing power can be purchased by ~anyone and most of those people are not specialists and have essentially no ability to write software for the platform, so (from a design perspective) it's a waste of time designing an interface that allows them to do so.

If your goal is to have a small powerful platform aimed at people who want to run software on it directly that's completely doable. Just see the raspberry pi and the raft of similar devices now available.

Closi2• 6 years ago

The engineering that went into this block was probably in the thousands of person-hours and involved some pretty rare engineering skills.

It’s probably in the tens of thousands - one person working on it by themselves for a year gets you to 2000 hours.

baybal27• 6 years ago

No, even a one man team can do a cookie cutter SoC with hard macros these days.

Mid-tier SoCs of around 2010-2012 period were done by teams of just 20-30 for most of brands on the market.

Only after the mobile market skyrocketed did RnD expenses skyrocket too as market competition started to get serious.

Closi2• 6 years ago

I’m assuming that this one wasn’t just a one man band, doing their own QA and making their own engineering samples.

Looks like the author updated their hours estimate.

rahimiali• 6 years ago

1) the age (now?) says “hundreds of thousands”, up two orders of magnitude from this. 2) it’s referring to the WiFi IP core, not the SoC itself, which seems an appropriate estimate. Just reading the 802.11 standard would account for a chunk of that time.

dang• 6 years ago

Related from 5 days ago: https://news.ycombinator.com/item?id=25742276

rmoriz• 6 years ago

I wish we would be able to repurpose such closed down devices to run free software.

solarkraft• 6 years ago

Yep. It's easy to imagine how many of these will be electronic waste in not that many years.

mhh__• 6 years ago

It's easy to imagine how many of these will be electronic waste in not that many years.

I'm pretty in favour of requiring some level of "the user can run whatever they want, supported but at their own risk" for every (reasonable) device for exactly that reason.

PragmaticPulp• 6 years ago

Even if we could, in 2021 it's far easier to pick up a more powerful, more appropriate SBC and simply use that for a project. A Raspberry Pi 4 is only $10 more and immensely more capable, as well as well-supported and heavily documented.

Unfortunately, it's not in manufacturers' best interests to make these devices hackable. All costs considered, Amazon is likely selling these near at-cost. They're not selling the hardware as an independent product. They're selling it as an extension to their larger platform (Alexa). As soon as a product becomes hackable, some portion of sales are lost to enthusiasts who buy one thinking they might hack it some day. If this gets out of control, eventually the company is subsidizing enthusiasts and the whole proposition becomes a net loss for the company.

Jon_Lowtek• 6 years ago

The whole thing might even run at a net loss with corporate arguing it is an investment.

Amazon knows voice agents are on track to become a primary human-computer interface, a user agent for the internet, and Amazon wants to own a big share of that pie, because that is a gatekeeper tech. Those who own it can ensure people buy on their marketplace, and companies that want to sell to these consumers must pay a fee. Amazon can't reasonably compete with google or apple in the "voice agent on the smartphone" segment until some antitrust forces those to offer a choice of agents, so they do the smart thing: push hard for a quasi monopoly on ambient microphones.

Over time there could be an antitrust to tell Amazon they must open "their" voice agent infrastructure for the competition, and they will use this "investment" as an argument why that would be unfair to them. Even if, they will rather add an API limited to accredited competition instead of giving consumers root on the hardware.

The whole concept is hostile to the consumer by design: imagine a keyboard would include a non-replaceable typing agent that can only type on websites that are partners with the keyboard manufacturer (as in "buy the skill for your keyboard to type on HN") and all your inputs are first sent to the manufacturers servers. Hey you get auto-complete and the keyboard is cheap. That is the state of voice agents.

system2• 6 years ago

Isn't Arduino enough?

syntaxing• 6 years ago

It’s super impressive that someone can just look at a die image and know what blocks are what. Kinda wished I knew more to about ASICs to be able to do the same.

intricatedetail• 6 years ago

Is that extra Wi-Fi being unused? Not difficult to imagine that this could have been a tap for LE to listen in if someone turned off WiFi

monocasa• 6 years ago

It doesn't seem to be hooked up to the antennas, so it seems fairly innocuous.

astrange3• 6 years ago

If you wanted to do that you wouldn’t need two wifi chips, one is enough if it’s not going to listen to you turning it off.

MJalali• 6 years ago

I always enjoy to see the details of boards

Amazon Echo Flex Teardown: The Big Silicon

Related Stories