Hi Spanner,

That's a tiny size! And standard sized pins for ease of use without the Arduino header spacing difficulty.

This, and the Genuino 101 seem like true modern-day successors for original Arduino : )

What would be cool would be underlying code/RTOS that would allow uploading it directly via BTLE from

your phone. Deploy it outdoors, and be able to upgrade the application without having to plug into it basically! : )

Until Intel publish proper data sheets for this and related processors the whole thing remains a huge puffed up PR campaign like "Intel Inside" but with less detail. I've looked a few times and there just isn't a proper Quark datasheet. What we do know is that its an 8086 based architecture clocked at  a rather slow 32MHz with only 80k of RAM and 384 k of flash.

So that's about as much welly as an early ST Cortex ARM M3 micro.

Intel will need to do a lot more (including publishing data !!!!) to even slightly dent the ARM grip on the small embedded systems market.

MK

Thanks for the link - what a truly dreadful website - lots of links to ancient (and not all free to download) academic papers - some insane demo files that want to install some weird code on your computer (haven't they heard of video ?) - but no decent explanation of what the technology is and what it can do. So after 10 minutes on their website I still have no idea of what useful function the neuron processor on the Quark could actually handle - could it recognize a number plate or simulate a low pass filter - I have no idea.

I go back to my original point - this is new and weird stuff on a cheapish micro - they are going to have to explain what it does to sell any.

MK

Have a look here

https://github.com/01org/Intel-Pattern-Matching-Technology

Thanks Daniel but that's code - not documentation or a reasonable description as to how it works and what it can do.

To give you an example - look at how TI support their range of inductive sensing chips - much simpler devices and 100x more effort to explain them, document them and help potential users apply them.

Inductive Sensing | Inductance to Digital Converter | Technical Documents | Sensor Products | TI.com

MK

If you want good docs, or a company who has the slightest idea how to service normal customers (you are not Lenovo, Acer, Apple, Dell, nor HP, and you're not buying millions of Xeons for your data center... are you?), you should consider other products from other companies, maybe.

Intel does need to do a better job of documenting it but they do give a fair amount of additional information if you talk to them rather than just poking around their website and using the Arduino 101 stuff. 

There's a lot going on with the Curie module that hasn't fully made the rounds because it's buried in documentation that hasn't been available to the general public for long and isn't easy to locate even if you have it.

First is the Quark SE (I'm not certain if it's exactly the C1000 but it's closely related if not) that others have mentioned.  It alone has 3 cores -- an x86 (no FPU or HT), an ARC core (with single precision FP and some other DSP extensions -- bit like a low end Cortex-M4) and a neuromorphic pattern matching ASIC(ish) core licensed from General Vision based on their Neuromem CM1K but with fewer neurons.  The ARC is where most of your code runs if you use the Arduino tools.  The general idea is you do your IO/DSP stuff on the lower power consumption (but same clock) ARC and only wake the x86 for certain types of heavy lifting. These are all on a single die with a high speed bus interconnecting them along with the integrated shared memory and flash.  The memory specs bandied about are somewhat misleading as there's both shared memory and local memory for the different cores.

Second, there are 3 other chips in the module: a charge controller (I forget the brand but it's well known), a MEMS IMU chip from Bosch and a BLE chip from Nordic Semiconductor (nRF51, I think) packaged in the module along with the Quark SE.  It really is a module -- a thin PCB with multiple wafer level chips including one on the bottom between the two sets of solder pads.

The x86, ARC, IMU and nRF can all communicate with each other independently so you can selectively power down anything to conserve battery.  The IMU chip also lets you power down the gyro independently which is a good idea if you don't need it since it's most of the chip's power draw.

The Neuromem core is probably one of the most interesting things that separates it from any other solution -- though oddly it's barely mentioned and always talked about in very oblique terms by Intel.  General Vision is the best source of info on it.  Note, you can download a library from General Vision's website that works with the Arduino 101 if you want to play with the pattern matching tech and they have desktop software to help you design and deploy solutions for it.   General Vision uses terminology that makes it sound like a classic neural network but it's quite different than that.   Still, it can be incredibly powerful for similar sorts of tasks.  Intel also offers their own library for it if you use their Zephyr distro and license their IQ libraries.

There are a few things I wish they'd done differently.  First, it really needs a bit more memory -- if you use their Zephyr distro, there's hardly any left for your application.  Second, in some cases the height is a problem but more area might not be and breaking out the components yourself would cost a LOT more.  Third, I wish they'd used a later nRF chip -- one that is firmware upgradable to BLE 4.2.  Fourth, there's a bug that forces you to add some external components if you don't want it to drain your battery -- the low power cutoff in the power management setup was done incorrectly and it's a hardware issue not firmware.  Fifth, it's a bit tricky if not impossible to get full throughput over BLE with their current firmware.

Finally, the documentation is pretty light particularly for low level firmware developers and you are left to glean what you can from their Zephyr OS code.  I believe they are working on some of these things.

Anyway, it took awhile to dig through and figure all that out so I figured it was worth passing along here.

Intel does need to do a better job of documenting it but they do give a fair amount of additional information if you talk to them rather than just poking around their website and using the Arduino 101 stuff. 

There's a lot going on with the Curie module that hasn't fully made the rounds because it's buried in documentation that hasn't been available to the general public for long and isn't easy to locate even if you have it.

First is the Quark SE (I'm not certain if it's exactly the C1000 but it's closely related if not) that others have mentioned.  It alone has 3 cores -- an x86 (no FPU or HT), an ARC core (with single precision FP and some other DSP extensions -- bit like a low end Cortex-M4) and a neuromorphic pattern matching ASIC(ish) core licensed from General Vision based on their Neuromem CM1K but with fewer neurons.  The ARC is where most of your code runs if you use the Arduino tools.  The general idea is you do your IO/DSP stuff on the lower power consumption (but same clock) ARC and only wake the x86 for certain types of heavy lifting. These are all on a single die with a high speed bus interconnecting them along with the integrated shared memory and flash.  The memory specs bandied about are somewhat misleading as there's both shared memory and local memory for the different cores.

Second, there are 3 other chips in the module: a charge controller (I forget the brand but it's well known), a MEMS IMU chip from Bosch and a BLE chip from Nordic Semiconductor (nRF51, I think) packaged in the module along with the Quark SE.  It really is a module -- a thin PCB with multiple wafer level chips including one on the bottom between the two sets of solder pads.

The x86, ARC, IMU and nRF can all communicate with each other independently so you can selectively power down anything to conserve battery.  The IMU chip also lets you power down the gyro independently which is a good idea if you don't need it since it's most of the chip's power draw.

The Neuromem core is probably one of the most interesting things that separates it from any other solution -- though oddly it's barely mentioned and always talked about in very oblique terms by Intel.  General Vision is the best source of info on it.  Note, you can download a library from General Vision's website that works with the Arduino 101 if you want to play with the pattern matching tech and they have desktop software to help you design and deploy solutions for it.   General Vision uses terminology that makes it sound like a classic neural network but it's quite different than that.   Still, it can be incredibly powerful for similar sorts of tasks.  Intel also offers their own library for it if you use their Zephyr distro and license their IQ libraries.

There are a few things I wish they'd done differently.  First, it really needs a bit more memory -- if you use their Zephyr distro, there's hardly any left for your application.  Second, in some cases the height is a problem but more area might not be and breaking out the components yourself would cost a LOT more.  Third, I wish they'd used a later nRF chip -- one that is firmware upgradable to BLE 4.2.  Fourth, there's a bug that forces you to add some external components if you don't want it to drain your battery -- the low power cutoff in the power management setup was done incorrectly and it's a hardware issue not firmware.  Fifth, it's a bit tricky if not impossible to get full throughput over BLE with their current firmware.

Finally, the documentation is pretty light particularly for low level firmware developers and you are left to glean what you can from their Zephyr OS code.  I believe they are working on some of these things.

Anyway, it took awhile to dig through and figure all that out so I figured it was worth passing along here.