I don't understand the real value in this. They take a bunch of BGA parts + passives and put the dies in a bigger BGA that approaches the same PCB area?
If someone has the manufacturing capability to put down one integrated BGA part, then they have the capability of putting down the DRAM and processors just the same as well as the 0201s or whatever passives to support it.
Only thing it seems to do is enable quicker prototyping by avoiding having to get things like the DDR memory traces correct. And a shorter BOM.
On the flip side, single source supplier, fewer memory options, fewer chip configuration options, etc. Places with the manufacturing capability of BGAs and 0201 probably aren't distracted much by 50(guess?) additional BOM line items and the reels of parts to realize that.
Also, how many hackers are actually putting down BGAs like this when you can buy a properly routed, tested and assembled BeagleBone for so cheap? I know a few are, but they are the 0.1% perhaps? All the makers are rocking Arduinos and Raspberry Pis and can't even spell B-G-A.
I was initially intrigued thinking it was considerably smaller, but it doesn't appear so. [0][1]
> I don't understand the real value in this. They take a bunch of BGA parts + passives and put the dies in a bigger BGA that approaches the same PCB area?
It's like DRY (don't repeat yourself) for embedded systems. This package wraps the common layout tasks of an embedded systems design in to an easy, pre-packaged module. The PCB designer no longer has to do the layout work of all of the DRAM and power-supply traces, which can consume a lot of time and effort when you're just trying to do a basic design.
> If someone has the manufacturing capability to put down one integrated BGA part, then they have the capability of putting down the DRAM and processors just the same as well as the 0201s or whatever passives to support it.
That's the other great thing about this package: The BGA pitch (distance between BGA balls) is a relatively large 1.27mm. This allows for the use of much cheaper PCB processes because the designer doesn't have to use very fine vias and traces to route in between the balls. For reference, many BGA packages these days have 0.8mm, 0.5mm, or even 0.4mm pitches and require more expensive PCB processes for the entire board just to place that one chip.
> Also, how many hackers are actually putting down BGAs like this when you can buy a properly routed, tested and assembled BeagleBone for so cheap? I know a few are, but they are the 0.1% perhaps? All the makers are rocking Arduinos and Raspberry Pis and can't even spell B-G-A.
Placing BGAs is quite easy for the hobbyist these days, especially now that hot-air stations can be had cheaply. You're right though, in that this product isn't targeted at the weekend DIYer. It's for people building boards who want a CPU but don't want to mess with tight PCB tolerances and all of the high-speed routing required. It also represents a very easy path toward manufacturing for those DIYers who do start with a BBB-based design and want to move to something they can mass produce and sell.
With this, they could drop a single, easy BGA package on their PCB and skip straight to doing design and layout only for the parts of their design that differ from every other embedded systems design (e.g. not the DRAM <-> CPU interface, not the power supply layout, and so on).
Getting DDR right is huge. It also means I can place this part on a 2 or 4 layer PCB, which is way cheaper than the 8 layers I need for a typical DDR layout.
> I don't understand the real value in this. They take a bunch of BGA parts + passives and put the dies in a bigger BGA that approaches the same PCB area?
> (...)
> Only thing it seems to do is enable quicker prototyping by avoiding having to get things like the DDR memory traces correct. And a shorter BOM.
My guess is that this sort of bundling may help in terms of driver support, as it provides a very specific bundle of hardware components and features.
Perhaps someone knowledgeable on the subject can chime in on the subject.
"have hundreds of passive components sprinkled around the board"
Resistors, inductors and capacitors are used to match the characteristic interface impedance (RLC) for many types of interfaces - there's often a integrated matching component near your Ethernet port. Capacitors are also used to "quiet" the power supply lines as they enter the integrated circuits.
These passive components are rarely used when interconnecting the different "chips" in a system as you design your system to use compatible parts. And with additional integration in the SoC, I'm guessing you'll need additional capacitors around the chip.
In short (too late?), I'm skeptical as to whether it will reduce the number of passive components at all. It will definitely make laying out a circuit board easier (except with BGA, you'll need multi-layer anyway).
SOMs need large connectors and, as I've found out recently, the DIMM connectors radiate a ton of EMI at LCD pixel clock speeds. Getting a proper heat sink is mechanically challenging too.
At this point I'd rather have the BGA that I can place on a low-speed PCB.
This fills a market that is currently only served with low-cost SOMs. The company I work for could directly replace the AM3354-based SOM that I write software for with this chip, especially if it brings out the GPMC, MMC, and a couple UARTs.
Does anyone know how the pinmux works on this part?
Shrug. It's okay, but there's still a lot of passives floating around on that reference design.
By contrast, some of the Freescale Cortex parts don't even need crystals to run USB (they use the USB signalling to tighten up the on-chip PLL/VCO). And they don't need capacitors on the crystal pins. So, you can use one of those Cortex parts with basically a USB connector, a switching power supply chip, some bypass caps AND THAT'S IT.
I also really don't understand why nobody integrates the ethernet PHY at this point. It would change from 20 pins to something like 5 pins.
Microchip has had a PIC18F97J60 for years. Why is nobody else doing this in the more advanced SoC's?
As of today, there isn't much of an advantage of doing a SiP for a chip like the AM335x other than the high speed design. Most of market is still with PCB SoMs for larger cortex-A chips.
The real advantage of SiP is with BLE or similar. That's because not only do they take away the complexity of RF design, and regulatory certification i.e FCC, IC, CE etc.. The market for BLE SiP modules that integrate the crystal, Balun, Antenna and are are certified is exploding.
Just a year ago, the cost difference between a BLE module and BLE chip was large enough to justify doing a custom design for anything more than 5K Units/Year. However, the Math is slowing moving towards higher and higher volumes. Already anything that ships less than 20K-30K per year doesn't justify the cost of doing a custom RF design.
Maybe in a couple of years, there will be a mass market for Beaglebone SiPs. Just not yet.
The real exciting part about this (to me) is the potential for a full fledged linux altoids tin computer. Suddenly the something like the Pyra/Open Pandora is a lot more realistic. Or even a (usable) hobbyist diy phone.
16 comments
[ 3.6 ms ] story [ 15.3 ms ] threadIf someone has the manufacturing capability to put down one integrated BGA part, then they have the capability of putting down the DRAM and processors just the same as well as the 0201s or whatever passives to support it.
Only thing it seems to do is enable quicker prototyping by avoiding having to get things like the DDR memory traces correct. And a shorter BOM.
On the flip side, single source supplier, fewer memory options, fewer chip configuration options, etc. Places with the manufacturing capability of BGAs and 0201 probably aren't distracted much by 50(guess?) additional BOM line items and the reels of parts to realize that.
Also, how many hackers are actually putting down BGAs like this when you can buy a properly routed, tested and assembled BeagleBone for so cheap? I know a few are, but they are the 0.1% perhaps? All the makers are rocking Arduinos and Raspberry Pis and can't even spell B-G-A.
I was initially intrigued thinking it was considerably smaller, but it doesn't appear so. [0][1]
Anyone have a different take on this?
[0] http://www.digikey.com/Web%20Export/techzone/microcontroller... [1] http://i0.wp.com/octavosystems.com/octavosystems.com/wp-cont...
It's like DRY (don't repeat yourself) for embedded systems. This package wraps the common layout tasks of an embedded systems design in to an easy, pre-packaged module. The PCB designer no longer has to do the layout work of all of the DRAM and power-supply traces, which can consume a lot of time and effort when you're just trying to do a basic design.
> If someone has the manufacturing capability to put down one integrated BGA part, then they have the capability of putting down the DRAM and processors just the same as well as the 0201s or whatever passives to support it.
That's the other great thing about this package: The BGA pitch (distance between BGA balls) is a relatively large 1.27mm. This allows for the use of much cheaper PCB processes because the designer doesn't have to use very fine vias and traces to route in between the balls. For reference, many BGA packages these days have 0.8mm, 0.5mm, or even 0.4mm pitches and require more expensive PCB processes for the entire board just to place that one chip.
> Also, how many hackers are actually putting down BGAs like this when you can buy a properly routed, tested and assembled BeagleBone for so cheap? I know a few are, but they are the 0.1% perhaps? All the makers are rocking Arduinos and Raspberry Pis and can't even spell B-G-A.
Placing BGAs is quite easy for the hobbyist these days, especially now that hot-air stations can be had cheaply. You're right though, in that this product isn't targeted at the weekend DIYer. It's for people building boards who want a CPU but don't want to mess with tight PCB tolerances and all of the high-speed routing required. It also represents a very easy path toward manufacturing for those DIYers who do start with a BBB-based design and want to move to something they can mass produce and sell.
With this, they could drop a single, easy BGA package on their PCB and skip straight to doing design and layout only for the parts of their design that differ from every other embedded systems design (e.g. not the DRAM <-> CPU interface, not the power supply layout, and so on).
> (...)
> Only thing it seems to do is enable quicker prototyping by avoiding having to get things like the DDR memory traces correct. And a shorter BOM.
My guess is that this sort of bundling may help in terms of driver support, as it provides a very specific bundle of hardware components and features.
Perhaps someone knowledgeable on the subject can chime in on the subject.
Resistors, inductors and capacitors are used to match the characteristic interface impedance (RLC) for many types of interfaces - there's often a integrated matching component near your Ethernet port. Capacitors are also used to "quiet" the power supply lines as they enter the integrated circuits.
These passive components are rarely used when interconnecting the different "chips" in a system as you design your system to use compatible parts. And with additional integration in the SoC, I'm guessing you'll need additional capacitors around the chip.
In short (too late?), I'm skeptical as to whether it will reduce the number of passive components at all. It will definitely make laying out a circuit board easier (except with BGA, you'll need multi-layer anyway).
At this point I'd rather have the BGA that I can place on a low-speed PCB.
Does anyone know how the pinmux works on this part?
By contrast, some of the Freescale Cortex parts don't even need crystals to run USB (they use the USB signalling to tighten up the on-chip PLL/VCO). And they don't need capacitors on the crystal pins. So, you can use one of those Cortex parts with basically a USB connector, a switching power supply chip, some bypass caps AND THAT'S IT.
I also really don't understand why nobody integrates the ethernet PHY at this point. It would change from 20 pins to something like 5 pins.
Microchip has had a PIC18F97J60 for years. Why is nobody else doing this in the more advanced SoC's?
http://www.myirtech.com/list.asp?id=467
http://phytec.com/products/system-on-modules/phycore/am335x/...
You can buy these with any level of high-density connector you desire, if you need all those pins. No need for a big messy BGA device.
The real advantage of SiP is with BLE or similar. That's because not only do they take away the complexity of RF design, and regulatory certification i.e FCC, IC, CE etc.. The market for BLE SiP modules that integrate the crystal, Balun, Antenna and are are certified is exploding.
Just a year ago, the cost difference between a BLE module and BLE chip was large enough to justify doing a custom design for anything more than 5K Units/Year. However, the Math is slowing moving towards higher and higher volumes. Already anything that ships less than 20K-30K per year doesn't justify the cost of doing a custom RF design.
Maybe in a couple of years, there will be a mass market for Beaglebone SiPs. Just not yet.