6502 or ARM core selection?

BigEd · Post by **BigEd** » Mon Jan 02, 2012 10:42 pm

Looks like there's an on-chip PLL - I think this is normal, so you don't have high frequencies on the board. The clock in is 18.432MHz, so the PLL is presumably configured for 10x, or maybe for 9.76x

Dajgoro · Post by **Dajgoro** » Mon Jan 02, 2012 10:54 pm

I don't understand how a clock frequency can be multiplied(wiki didn't help). And what about the data bus, how do i build it so that there would not be any major problems, like crosstalk?

BigEd · Post by **BigEd** » Mon Jan 02, 2012 11:31 pm

If you imagine a frequency locked loop, that's fairly easy to understand: you have a fast oscillator running at some suitably high but approximate frequency, you divide it down by 10 and compare it to the input clock, and you adjust the fast oscillator until the comparison is equal. Now it's running at 10x the input rate.

A phase locked loop is the same, but more subtle.

Dajgoro · Post by **Dajgoro** » Tue Jan 03, 2012 1:44 am

That means that in the cpu itself is a very fast oscillator, and it only uses the slower crystal clock for comparison?

Anyway any ideas for the pcb?

GARTHWILSON · Post by **GARTHWILSON** » Tue Jan 03, 2012 1:57 am

Quote:

Almost always when i build something new, it does not work on its first powerup, or after a while i decide to rewire some stuff, because i found out that there are better ways of doing things... So lets say i make a schematic for a sbc, and then i order it, and i solder all the components, and then i want to make changes and try just those changes to see how they perform, i would have to order a new pcb. I can't afford to order so many pcb-s.

A major problem is often layout errors in the PCB, things that don't match the schematic. Even when using OrCAD at my last place of work, the CAD operator who took our schematics, entered them on the computer, had us check the schematic, and then laid out boards with the CAD doing the rat's nest with the netlist and so on, still made layout errors, even though the CAD was supposedly there partly to eliminate that problem. A new engineering manager introduced us to a way to check layouts manually that was more foolproof, and I've been using it ever since, and I introduced it to another engineer who also admitted later that it was way better. Both of us were kind of forced into it kicking and screaming on the inside.

What you do is to have the PCB design in CAD in front of you, in a copy that you can destroy as you go. It's best to have two people on this (one on a paper copy of the schematic and one on the CAD), but I normally don't have any help. You pick a net on the schematic, follow it on the CAD, marking the schematic with a colored pencil as you check that net on the CAD, then when you're satisfied that the net goes to everything it should on the layout and not to anything it should not go to, delete that net on the CAD. Now it's deleted on the CAD and marked on the paper schematic. If you find an error, write it down, or fix it in another window. As parts get all the nets they're connected to checked and deleted, delete the part also. If there are errors, they will show up. You might find for example that you're left with a line on the schematic that's not marked and there's nothing left on the board, because that line did not get routed.

I don't like the way any of the CADs work for schematics, although I got very proficient at OrCAD. For example, schematics can be made the neatest and most readable (because of the fewest corners, etc.) if you can re-arrange and re-space the pins of an IC every time you draw it on a schematic; but the CAD doesn't let you do that. Anyway, to this day I still do my schematics by hand. I check my layouts with the method above, and my boards are right the first time, every time. The CAD does have DRC (design rules checking) but it doesn't understand the unorthodox things I do to get extreme density and it will give hundreds, if not thousands, of things it thinks are design rules violations, but if I go down the list and check them all, they're all ok.

Quote:

2. I don't have the fancy pcb, i could find a way to get a normal single layer pcb.

You can usually get much better performance with two layers than one, and you might find that the cost is not much different. Certainly you can get better density, and then you don't have to pay for as many square inches (or centimeters). The connection integrity is much better if the holes are thru-plated, and after that, if there's only one layer instead of two, the second side starts out to be all copper and just gets etched away anyway. Going beyond two-layer is where the cost starts going up more because they have to make multiple thinner boards and then laminate them together.

Quote:

3. Why [do] all modern computers have coils in the power supply?

They are switching regulators, to get much better efficiency. In the case of power supplies that are powered from the 120/240V 50/60Hz mains, a switching regulator also helps reduce weight, and avoid putting so much noise on the input power line. If the voltage starts out lower than what you need (like if you need 5V or 12V from two 1.5V batteries), you can't get there without reactive components. There are switched-capacitor circuits to step voltage up or invert it, but what you can do with that is rather limited. I have designed and used several switching regulators, although mostly (not entirely) starting with off-the-shelf switching-regulator controller ICs. Instead of trying to explain the operation here, I'll refer you to http://en.wikipedia.org/wiki/Switching_regulator or other articles a web search will turn up.

GARTHWILSON · Post by **GARTHWILSON** » Tue Jan 03, 2012 2:02 am

Quote:

That means that in the cpu itself is a very fast oscillator, and it only uses the slower crystal clock for comparison?

Basically, yes. The faster oscillator is usually a voltage-controlled oscillator (VCO). Its output is divided down and compared to the crystal oscillator frequency (which may also be divided down), and an error voltage is fed to the VCO to correct it to the desired frequency. You won't find crystals in the hundreds of MHz or GHz.

Dajgoro · Post by **Dajgoro** » Tue Jan 03, 2012 2:23 am

GARTHWILSON wrote:

Quote:

3. Why all modern computers have coils in the power supply?

They are switching regulators, to get much better efficiency. In the case of power supplies that are powered from the 120/240V 50/60Hz mains, a switching regulator also helps reduce weight, and avoid putting so much noise on the input power line. If the voltage starts out lower than what you need (like if you need 5V or 12V from two 1.5V batteries), you can't get there without reactive components. There are switched-capacitor circuits to step voltage up or invert it, but what you can do with that is rather limited. I have designed and used several switching regulators, although mostly (not entirely) starting with off-the-shelf switching-regulator controller ICs. Instead of trying to explain the operation here, I'll refer you to http://en.wikipedia.org/wiki/Switching_regulator or other articles a web search will turn up.

I know how switching power supply works, i forgot to mention on which coils i was referring to. I saw in lots of schematics coils located near the cpu itself, and in the voltage regulator circuit.

Edit:

Could i use this http://www.ebay.com/itm/1-Pcs-Copper-Cl ... 256b0b4d21 and what would be the best way of etching it?
I could also use this: http://www.ebay.com/itm/2-x-TQFP-100-TQ ... 5d31b0356a

GARTHWILSON · Post by **GARTHWILSON** » Tue Jan 03, 2012 9:17 am

I was never very successful at making my own PC boards. I tried the etch-resist pens but the etchant would eat right through them. I tried painting the lines with a very thin brush, and although it worked, I could not make narrow lines. After etching, I soaked and scrubbed to get all the etchant out of the corners where the coper traces met the board, but still got corrosion sometimes after a few months because apparently I didn't get it all out. Where I worked in the mid-1980's and we did wide stripline transmission lines on PC boards for RF power amplifiers, we would cut rubiliths, put a solution on the board and spin it off in the darkroom, then put the rubilith over it and expose it to UV light, then develop it and etch it, and that worked pretty well, but I don't know where to get whatever it was we used, and I know I still couldn't get very narrow lines. A good PC board house can make traces basically no wider than the thickness of the copper, and if you look at them under a microscope, they truly look square. It's amazing. I'm not saying we go that narrow at our company now, but I regularly do .007" traces and they've proven to be 100% reliable. We go down to .008" vias on the smaller .031"-thick boards, and the thru-plating on them has also proven to be 100% reliable. In the 80's, before we had CAD, we use Bishop Graphics patterns and crepe tape we stuck down on dimensionally stable film, usually 2X or 4X actual size, then take them to a graphic arts shop to have them reduced for the PC board house. At that time however, we didn't have any lead spacing closer than .100".

BigEd · Post by **BigEd** » Tue Jan 03, 2012 6:54 pm

New thread here for the etching of PCBs.

And here for high-speed design - for questions about crosstalk and so on.

Cheers
Ed

BigEd · Post by **BigEd** » Mon Jan 16, 2012 9:43 am

BigEd wrote:

Another challenge is that for linux you'll need at least 128M RAM (maybe 32M will be technically possible. See Tinycorelinux FAQ. Maybe even 16M - see dd-wrt supported devices)

Allow me to correct myself: here's a project putting linux on a cheap (£10) ARM-based photoframe which has only 8M of RAM & 32M of flash. (They've been able to upgrade the RAM in-place to 32M, but that wasn't necessary for linux.) (Via Hackaday)

Cheers
Ed

ElEctric_EyE · Post by **ElEctric_EyE** » Mon Jan 16, 2012 4:14 pm

wow, it is fascinating that that is the smallest PC running Linux. That is an awesome hack!

Dajgoro · Post by **Dajgoro** » Thu Mar 01, 2012 5:20 pm

The mythical Raspberry pi is finally out for sale!
I immediately wanted to buy two boards(one for my friend), but then i noticed that on the country list Croatia was missing!
I sent a email to someone at Raspberry pi, but no answer.
Does anyone have some info how could it be possible to buy one?

cjb · Post by **cjb** » Thu Mar 01, 2012 6:00 pm

BigEd wrote:

Another challenge is that for linux you'll need at least 128M RAM (maybe 32M will be technically possible.

lol what.
http://kildall.apana.org.au/~cjb/sable1.jpg
http://kildall.apana.org.au/~cjb/sable2.jpg

That's in 2 megabytes of RAM. The OpenWRT-based systems are horribly bloated and don't do anything smart with memory at all.

BigEd · Post by **BigEd** » Thu Mar 01, 2012 6:18 pm

I salute your 2Mbytes (with lots of swap.) I've used compcache with some success on my XO-1 - it swaps to RAM with compression. Makes the RAM go further.

BigDumbDinosaur · Post by **BigDumbDinosaur** » Sat Mar 03, 2012 6:00 am

GARTHWILSON wrote:

What you do is to have the PCB design in CAD in front of you, in a copy that you can destroy as you go. It's best to have two people on this (one on a paper copy of the schematic and one on the CAD), but I normally don't have any help. You pick a net on the schematic, follow it on the CAD, marking the schematic with a colored pencil as you check that net on the CAD, then when you're satisfied that the net goes to everying thing it should on the layout and not to anything it should not go to, delete that net on the CAD. Now it's deleted on the CAD and marked on the paper schematic. If you find an error, write it down, or fix it in another window. As parts get all the nets they're connected to checked and deleted, delete the part also. If there are errors, they will show up. You might find for example that you're left with a line on the schematic that's not marked and there's nothing left on the board, because that line did not get routed.

That's the method I've used for a long time to weed out layout errors. It's quite effective and I recommend it to anyone trying to lay out a PCB. It sure beats throwing away PCBs that are discoverd to have an intractable error after they have been made and paid for.

Quote:

Going beyond two-layer is where the cost starts going up more because they have to make multiple thinner boards and then laminate them together.

The cost differential is coming down and square inch for square inch, a four-layer board is about 60 percent the size of an equivalent two-layer design. The advantages go well beyond physical size. Four-layer designs, in which the inner layers handle power and ground, are potentially a lot less noisy, generally allow increased component density (you're not having to use up space for power and ground runs) and greatly reduce voltage fluctuations and ground bounce, the latter which sabotages a lot of designs. I did my POC unit on a four-layer board and it proved to be stable at 12.5 MHz (maximum speed the I/O hardware would tolerate). In fact, in a fairly recent test, I removed all the decoupling capacitors and it still was stable at 12.5 MHz. That would be attributable to the fact that the inner power and ground layers act as a giant bypass capacitor and were able to sink most of the switching noise. However, don't take this to mean you should forego decoupling caps. Any good design will have one for each active device on the board.