Amazon Echo Flex Teardown part1
Vložit
- čas přidán 29. 12. 2020
- Extreme teardown of the Amazon echo flex. In part one we look at the overall assembly and then study the MEMs microphone, ADC convertor and the main SOC. Video include decap of the silicon.
Photos may be found at my blog:
electronupdate.blogspot.com/20... - Věda a technologie
I know this is very niche interest, but for those of us who are interested, this is golden and we really appreciate it!
I love these videos. Any time I see electronupdate I click instantly
Would love details on the "hardware mute" functionality.
This.
I would guess something simple like holding the ADC in reset.
Using a dedicated speakerphone chipset that likely is just a small signal MOSFET shorting the ADC input pin to analogue ground. either part of the chip or an external part.
Just checked information from Mediatek, the SOC only has 802.11bgn and BLE4.2, while the seperate WiFi-Chip supports 802.11ac and Bluetooth 5.0. Seems Amazon wanted or needed a faster wireless connection than the SoC could offer and it was easier/cheaper to use a dedicated WiFi chip instead of using a different SoC.
Man, your microscope photographs are amazing!
glad that you still are ok and alive electronupdate
These in-depth teardowns are greatly appreciated, keep them coming. Happy new year!
That thing has a lot to unpack. Think they make money on it upfront or is it a "Handle is free, the razors'll cost ya tho" situation?
100% the latter. Literally the main reason for this device's existence is to make people buy stuff through Amazon.
@@Gameboygenius Lord Bezos knows no bounds.
My guess would be that people start placing them into rooms where they "think" they don't need a bigger echo like bathrooms or a basement and a few months later they upgrade. In the long run all devices also help them to improve their voice recognition technology which could be interesting in the future for the automotive industry or other areas. Getting as many devices out as possible and as many users as possible (to train the system with all kinds of dialects, slangs, age groups etc) is the best thing to improve it quickly.
Absolutely fantastic. Can't wait for part 2. Welcome back.
Fascinating! I have to believe Amazon loses money on each and every one of these sold.
Hey Adrian! It's amazing how the tech crowd always watches the same content. I'm expecting Big Clive to comment here as well. We're a big technical family.
Cheers.
This is soo cool what you do here. I really apreciate that you share this kind of information.
Thank you for another amazing view of modern tech. Can't wait for the next one.
Great video! I'm looking forward to the next one.
wish you had scale in every slide
Banana cells for scale ;-)
Was going to write the same, curious about the actual die sizes
Love the narrated walkthrough of both the boards and silicon noting features and why things were possibly done a certain way. Excellent insights. Thanks for sharing.
so good! love the detail
Great video as always 👍
Thanks for sharing your experience with all of us 👍😊
Happy new year to you and your family 👍💖🎉
Happy new year, dude!
Yeah that SOC is a lot more complex than the usual ancient ICs still in use which are just about understandable. Would be more than happy with the quick layout overview as seen from the metal layer. Regards.
Super cool. I'm also interested in hardware mute, if any, as well as what the other little ic's on the main board are, up at the top around the adc.
Wonder if the dsp on that adc is beefy enough to do the microphone array processing or if that gets handed off to the main AP.
Happy New Year!!
Hi, just found your channel so I dont know if you have done it in the past but could you create a video detailing the decapsulation process of the ICs?
Glad to see a new video! I hope you're doing alright. It is odd to have a separate wifi/BT chip when it's built in already.
Probably done to offload processing to a separate chip, getting the core that would otherwise be dedicated to it free for on chip work or power management. That way they can have the main cores in a power down mode most of the time, and only one core with a reduced clock running to poll the wireless interfaces and the ADC, looking for either a MAC match or a sound level above the trigger level, which then powers up the other 3 cores at full speed to process the input. As well for playing audio all you need is a slow core to do the decoding, keeping power use low again, though it is likely using the ADC chip speakerphone DSP to subtract the output from the microphone, to still be able to respond to audio triggers as well.
As they need a Bluetooth stack in addition to wireless the chip with both functions in it was cheaper overall, and can use the same antenna units as well. Also easier to upgrade the unit if they need different wireless functionality or worldwide coverage having a chip that you simply either reflash or change a configuration register set.
@@SeanBZA That makes sense to me. Thanks for the detailed reply.
Is there any porting for the microphone inputs in the case or are they just direct to the sensors?
Haven’t seen one of your videos lately. Glad to see it. I would like to see more on that DC DC converted in the mic. Is it something like an RC oscillator and magnetic?
Also hardware mute would be interesting. I suspect most of the silicon is going to be too small for much analysis. It’s the downside to cutting edge tech. Usually ancillary devices lag a little behind core devices but I guess we are at that point where value devices can have ultra small features.
Also iDK if this would be of interest but if you send me silicon I could put it in our SEM and send you the images (no charge)
The pictures of the circuit boards are in an amazing quality. What kind of camera-equipment do you use to take these?
Whats the black plastic part?
Hello, my idea is to incorporate the echo flex inside the jbl flip 4, for this I need to power the echo with 3.7v provided by the jbl battery, surely somewhere the 5V of the echo is reduced to join those parts, then perform the internal audio connection and test how both microphones work without drilling holes in the jbl's cabinet, what do you think? The video is very interesting. Greetings
nice!
I imagine the separate diaphragms in the MEMs microphone are not bonded out individually because they’re internally connected in parallel, the reason for two diaphragms is to get a better fill factor and better frequency response, the design of the grid is also key to frequency response. If you still have it could be interesting to cut it out and look at the other side? Should be a similar pattern.
That separate WiFi IC is weird, the onboard WiFi of the MT8516 is only 2.4 GHz whereas the MT7658 does 2x2 2.4 GHz & 5 GHz although no DFS support nor mesh peer-to-peer support so I’m not really sure why they would do this especially when trying to hit such an insanely low price point for such a complex device. Speculation: I wouldn’t be surprised if a key reason to support 5 GHz is to collect beacon and probe frames and their RSSI of nearby 5 GHz devices like your phone to tell when you’re moving around. I'm sure they would love to feed that data into their magic AI.
Definitely take an in depth look at the “button that electronically disconnects the mics”.
Upon further research the only other thing I can find that might need a separate WiFi & Bluetooth IC is this "Amazon Sidewalk" thing but there's very little info about it.
Really curious why only one sensor in the MEMS is used
I have two theories:
1. It's possible they build a stereo version of this sensor, I feel like the separation would be terrible, but I guess it's possible if you want to drive a stereo input. In this case it may be more beneficial to make only one chip and package it differently based on the application.
2. Maybe they defect rate is high enough that they make a dual sensor, but end up with one of the units on the chip broken most of the time. So they can just bond whichever side is working correctly.
It seems like one of these, or a combination of the two, could be very possible explanations.
Looks like Mediatek sells a develpoment kit for integrating with Amazon Alexa using the same chip. I wonder if this is built on the same platform. www.mediatek.com/products/voiceAssistantDevices/mt8516-ffa
Where the hell is part 2? I'm very interested in the output of the power supply
6:31 A1 here might mean A = first silicon mask revision, 1 = first metal mask revision
Correction: 0 would be the first metal revision. 1 would be the second
It's very reasonable to believe that such a simple design could be shipped in the 1st revision.
what really is the point of having 2 microphones in each mems package for stereo output? they're separated by like.... a few microns. if even the highest audible frequencies have wavelengths on the order of centimeters then both of the mems microphones in the same package are always going to be experiencing the same pressure and producing the exact same output. this is why they only connect one microphone and then put another mems package on the other side of the board, because there probably is useful directionality etc. information that can be extracted at that scale. but stereo output from the same package? why not just fake it and double the signal electronically?
Maybe improved performance / gain at the vocal fq range with 2 mics per package.
2 microphones to increase sensitivity, as you now have double the surface area, also having 2 and only bonding one means you can have a single one fail chip scale testing and not make the whole unit useless, increasing yield, especially if you are not needing the most sensitive microphone. The MEMS sensor is basically a near 1GHz oscillator, using the mic capacitance to vary a LC tank, mixing it with another LC tank generated on chip, and then using the difference as the audio output. With them all consisting of silicon you can get the temperature tracking to be pretty close, though there is a DC offset, likely removed by a low pass filter in the chip, and applied to a reverse biased diode in the reference oscillator, to keep the drift in the reference oscillator the same as the sense one.
My thoughts are that it was included for either the increased sensitivity or to facilitate noise/bias compensation. Having a 2nd mic package nearby already probably allows them to get both from that while also increasing phase separation for the directional component.
I wonder how practical it would be to wirebond that 2nd mic to something more fun.
@@SeanBZA thanks for that....those RF inductors were giving me a headache. So bonding in the other microphone might change the LC frequency to something like 500MHz?
So the transformer in the center of the microphone IC is the mixer/downconverter presumably ....think there has to be a couple of diodes in there.
I think they dampen each other and prevent each other from oscillating on their own and getting into resonance in some frequencies.
Sort of like a difference amplifier, only signals that pass a certain criteria from each MEMS is accepted by the DSP.
Big Clive sent me.
"It's just amazing what I got for $15"
That's what they want you to think...but the data/information you will hand to them over the product's lifetime is worth far more than any of us could possibly imagine.
It also wouldn't be a surprise if these were secretly subsidized by a three letter agency.
Thermo-what material?
Thermal interface material.
@@Gameboygenius Perfect. Thanks!
Nice videos though I do think your audio/microphone could use work.
FYI, some of your audio edits cut off parts or even whole words at begging and/or end of the clips.
so dense!
Really enjoy the videos. Thanks for taking the time.
It looks like Amazon is selling this below BOM cost. I wonder if they could get in trouble for that.
$15 a loss leader? IMHO google just want one in every home. Very interesting looking forward to seeing under the silver (lifting its skirt 🤣) 2x 👍
I had to google up wth an Amazon Echo Flex is in the first place.