Electronic – LNA design question – series drain inductor

Often I have a question, but I don't even know enough about the topic to know how to use the right words to phrase the question. Is one of the following one of the questions you really intended to ask?

• Which is better, the Micro 2440 "Stamp Module" or the Micro 2440 "SDK-Board" ( both from FriendlyARM ), for running a program I custom-write to display stuff on a touch panel? Both. These two boards are designed to plug into each other; neither one is useful alone. (But perhaps the "Stamp Module" could plug into some carrier board other than the SDK board, as described below).

• Can I buy a bunch of FriendlyARM Mini 2440 SBC SDK-Board + Stamp Module and FriendlyARM accessories from FriendlyARM, install Linux on it, install my own custom application on that, put it in a box and sell them to customers? Yes.

• Are there standard ARM Linux boards, small TFT monitors, etc. kind of like the FriendlyARM Mini 2440 SBC, that I could buy as separate components, assemble, then put my own custom application on it and sell to customers? Yes. Some of them are listed at "motherboards that run Linux".

• Do I need to custom-design my own hardware from scratch? No. While you could do that, another option is to assemble separate components, as mentioned above.

• If none of the standard ARM Linux board was quite what I wanted, could I save money by hire someone to custom-design a handheld Linux board for me that has exactly what I need, rather than using these off-the-shelf modules that have a bunch of stuff I don't want? Maybe. Because of economies of scale, standard off-the-shelf components are going to be lower cost than custom-designed products produced in lower quantities. It's often lower total cost to use a OTS component rather than design something new, even when the OTS component includes several things you never use. (This is one reason why smaller, lighter products are sometimes more expensive than bigger, heavier products that do more).

• If the standard ARM Linux boards plus the accessories I need are too big to fit in the handheld shape I want, but only because they include a bunch of stuff I wasn't going to use anyway, could I hire someone to more-or-less copy the design, leave out the parts I don't use, and fit it into the shape I want? Yes.

• • Option 1: prototype with a pile of "open hardware" boards. Once you confirm that the prototype does everything you want it to do (except it's the wrong shape and weighs too much), you or any engineer you hire can copy-and-paste the original schematics into one big schematic, edit that big schematic to cut out all the stuff you don't need, (optionally) cut the schematic into multiple boards, and then lay out one or (optionally) several PCBs in the appropriate shape you need. The engineer can design the PCB so it connects the remaining components exactly the same way, so the software can't tell the difference.
• • Option 2: If it looks like a pile of non-open-hardware boards, all from the same manufacturer, will do what you want, you can ask that manufacturer to design a custom board (or multiple boards) that fits exactly the shape you want. Pretty much any manufacturer is more than willing to produce a customized version of their products, if you pay them enough money for the up-front design and per-board price. If you are lucky you can negotiate that price to something you can live with.
• • Option 3: prototype with a pile of boards, some of them "open hardware" and some of them non-open-hardware from a variety of manufacturers. The engineer designs a "carrier board" so your final product includes the carrier board and a bunch of off-the-shelf components plugged into it -- all of the non-open-hardware components, and perhaps some of the off-the-shelf open-hardware components. Option 3 often gives the fastest time-to-market, because the carrier board does so little that it's much quicker to design and test it than a huge monolithic board.

There are a number of architectures, actually. And they're not as different as claimed by the different vendors (although there are still differences.)

NVIDIA typically builds GPUs that work a lot like general-purpose CPUs: There's a large framebuffer out in GDRAM, there's a number of large caches for texture fetch and buffering, and there are groups of pixel processors (typically 4x2 pixels) that operate on some particular primitive at a time. For framebuffer blending or writing, the I/O goes through the memory bus, somewhat helped by the caches. If multiple different objects cover the same pixels (overdraw) then the same pixels may go to/from the framebuffer more than once during a particular frame.

POWERVR and Intel typically build "tiled" processors, where they have a similar architecture (4x4 tiles IIRC) but to avoid the big input/output load on the GDDR, they sort all the primitives ahead of rasterization, so they only need to load/write a particular block of framebuffer pixels once per frame. At least, that's their design goal -- this is not always necessarily achieved. The benefit is that you need to spend less bandwidth on reading and writing back framebuffer pixels. However, approximately the same amount of texture bandwidth is used, so the savings aren't necessarily orders of magnitude (it's scene dependent.)

The Xbox architecture is a bit of a hybrid -- it has EDRAM that's bigger than itty-bitty tiles, but smaller than a full HD framebuffer. The GPU runs through the primitives, and writes them to the EDRAM, which then "resolves" to the output framebuffer (this also calculates things like anti-aliasing.) The EDRAM framebuffer is made to be effectively fast enough to be non-blocking. The screen will be split into some number of tile areas -- 1 through 4 for a particular frame, typically.