65XX SBC general help and color display help needed

[Arlet]

One option to make it smaller is to move to SOIC packages. A lot of work, though, to redesign entire board.

[backspace119]

Arlet wrote:

One option to make it smaller is to move to SOIC packages. A lot of work, though, to redesign entire board.

I've tried very hard to stay away from surface mount. I have two aversions to it, the first being that I can't socket smd packages, and the second being that although I've soldered them before, I hate doing it, and if I don't have to get a stencil and paste (and figure out a reflow oven) that's always a plus.

The video card and audio card I have planned use VLSI chips that are surface mount, and I've actually considered having those boards assembled as well as fabbed, just to avoid having to solder them myself.

[GARTHWILSON]

several points about SOICs and SOJs, some plus and some minus:
SMT ICs with leads on .050" centers are not really difficult to solder (although as you say, you can't socket them); but SMT can be harder to troubleshoot. A plus for SMT ICs is that since the pins and their pads only occupy the component-side copper layer (rather than all layers like thru-hole does), routing in other layers can be easier, without all those holes in the way. It does mean you will need more vias, and vias take board space too; but if you're assembling by hand, vias can be put in the pads, unlike normal automated assembly where applying the solderpaste by silkscreen would make a mess on the other side as the solderpaste squishes through the hole. Without paying extra for super fine trace & space, you can only get one trace through between adjacent pads of SOIC or SOJ on the component side-copper layer (compared to three or four in thru-hole). On finer-pitch SMT ICs, you can't get any through. Now that's a pain!

[backspace119]

GARTHWILSON wrote:

several points about SOICs and SOJs, some plus and some minus:
SMT ICs with leads on .050" centers are not really difficult to solder (although as you say, you can't socket them); but SMT can be harder to troubleshoot. A plus for SMT ICs is that since the pins and their pads only occupy the component-side copper layer (rather than all layers like thru-hole does), routing in other layers can be easier, without all those holes in the way. It does mean you will need more vias, and vias take board space too; but if you're assembling by hand, vias can be put in the pads, unlike normal automated assembly where applying the solderpaste by silkscreen would make a mess on the other side as the solderpaste squishes through the hole. Without paying extra for super fine trace & space, you can only get one trace through between adjacent pads of SOIC or SOJ on the component side-copper layer (compared to three or four in thru-hole). On finer-pitch SMT ICs, you can't get any through. Now that's a pain!

Ya, that's actually another thing, I can only fit two traces through the DIP pins if I'm lucky (on one side of the board that is). I tink it's because my settings are 6mil/6mil, and I think the pads that I've selected for the chips are kind of fat. I assume with SMT style I'd still be able to fit one, so nbd, but I'm not sure if I'd put vias on the pads, since if I had a lot of SMT stuff I might go ahead and build a solder paste dispenser, which works kind of like the big machines do, and it might make a mess.

[Dr Jefyll]

@Chromatix, you're at least slightly off target in saying 74F devices are no longer being manufactured. Digikey still stocks (and lists as "active") a fair selection of 74F devices. That said, the list may be shrinking as time goes by. I fully agree with your other points, although the disappointing power and fanout restrictions may be an acceptable tradeoff in some situations. 74AHC really is the first choice, as you say.

Chromatix wrote:

About 74F series logic: it's pretty fast, yes, but it's *very* power hungry, on the order of 35mW per simple logic chip. It has typical TTL input levels, input loads (this is *not* a high-impedance input, due to the bipolar transistors involved) and fanout restrictions. It's also no longer being manufactured.

If you need fast logic, I recommend 74AHC series, switching to 74AC or 74HC for specific devices not available in that series. A quick pricing comparison suggests there's not much difference.

backspace119 wrote:

Thanks again for this, I just made the move and it saved me about 20mm on the y axis. [...] I'm going to keep looking around [...]

Happy to help. Maybe your next goal should be reducing the x axis. Look for stuff that protrudes, and see if it can be moved vertically or squished inward.

[Chromatix]

Granted, I only looked at one device, a hex inverter, taking it as characteristic of the whole series. Back in the day, I'm sure it would have been popular among the folks who needed high performance and didn't care much about power, such as minicomputer vendors, but it is *long* obsolete these days.

(For context, note that individual disk drives for a minicomputer often required hundreds of watts.)

[cbmeeks]

backspace119 wrote:

The problem is, OSHPark 4 layer boards are horrendously expensive ($10 per square inch, which at 45ish square inches right now, brings it to $450. You do get 3 boards with the order, but still).

I don't know if this has been mentioned, but I've used https://jlcpcb.com a couple of times and they have been really good. The quality was good and I literally got my boards within 7-8 days.

They are MUCH cheaper than OSHPark.

[backspace119]

Dr Jefyll wrote:

@Chromatix, you're at least slightly off target in saying 74F devices are no longer being manufactured. Digikey still stocks (and lists as "active") a fair selection of 74F devices. That said, the list may be shrinking as time goes by. I fully agree with your other points, although the disappointing power and fanout restrictions may be an acceptable tradeoff in some situations. 74AHC really is the first choice, as you say.

Chromatix wrote:

About 74F series logic: it's pretty fast, yes, but it's *very* power hungry, on the order of 35mW per simple logic chip. It has typical TTL input levels, input loads (this is *not* a high-impedance input, due to the bipolar transistors involved) and fanout restrictions. It's also no longer being manufactured.

If you need fast logic, I recommend 74AHC series, switching to 74AC or 74HC for specific devices not available in that series. A quick pricing comparison suggests there's not much difference.

backspace119 wrote:

Thanks again for this, I just made the move and it saved me about 20mm on the y axis. [...] I'm going to keep looking around [...]

Happy to help. Maybe your next goal should be reducing the x axis. Look for stuff that protrudes, and see if it can be moved vertically or squished inward.

Image3.gif

I actually looked at that last night, and I think I can push the 65SPI in, and some of the chips beside it, and get some x length off.

cbmeeks wrote:

backspace119 wrote:

The problem is, OSHPark 4 layer boards are horrendously expensive ($10 per square inch, which at 45ish square inches right now, brings it to $450. You do get 3 boards with the order, but still).

I don't know if this has been mentioned, but I've used https://jlcpcb.com a couple of times and they have been really good. The quality was good and I literally got my boards within 7-8 days.

They are MUCH cheaper than OSHPark.

Thanks for the link, I'll check them out too.

I have another question for everyone, while I was looking at clocks I got to thinking about a variable clock system, that could be changed by the computer itself. If I did this, I could tune the clock frequency with pretty decent accuracy, and squeeze every drop of power I can get out of the thing. This would also remove the need for having wait states, when I'm working with slow devices I can just slow down the clock to talk to them.

In particular, I was just looking at the 74LS628. It looks like an older device though, and I'm not sure if it's still relevant (also, it seems to only go to 30Mhz, which if I'm still running it through a flip flop, maxes it at 15Mhz, not bad, but I'd prefer to be able to try up to 18-20Mhz since some people have reported getting those speeds here).

I looked at programmable ones, but they all seem to be either tiny, too fast, only programmable once, or a combination of all 3. VCOs (like the 74LS628) seem to be a better option, but I've not looked at the other variable clock types yet.

[backspace119]

cbmeeks wrote:

backspace119 wrote:

The problem is, OSHPark 4 layer boards are horrendously expensive ($10 per square inch, which at 45ish square inches right now, brings it to $450. You do get 3 boards with the order, but still).

I don't know if this has been mentioned, but I've used https://jlcpcb.com a couple of times and they have been really good. The quality was good and I literally got my boards within 7-8 days.

They are MUCH cheaper than OSHPark.

Wow $107 for 20 boards! That's amazing! Thanks for this link, I'll definitely be going with these guys, I don't think even seeed can beat these prices.

[backspace119]

Here's the datasheet for the product mentioned in my last post: http://www.ti.com/lit/ds/symlink/sn74ls628.pdf

[Chromatix]

I suspect those particular devices were never made in CMOS versions. Broadly similar devices can be obtained under different part numbers, however.

My advice would be to avoid the complication of a continuously-variable clock source, and instead consider switching cleanly between certain chosen discrete clock sources. These can themselves, for flexibility, be integer divisions of a fast master clock; use a presettable counter and a JK flipflop wired to toggle every time the counter reaches zero (and reloads). I think there is a circuit for glitch-free clock source switching somewhere, or just stuff a new value into the preset register.

Usually, slow devices are accessed in a minority of cycles, so it makes performance sense to implement wait states to access them; all the other cycles are then still fast. Power saving, meanwhile, can be achieved through "race to sleep", assuming your static power consumption is low - that is, execute WAI and make your software interrupt driven.

Crystal oscillators take a while to start up and stabilise, and might themselves dominate your power consumption in some cases. In that case, having a slow oscillator that might be more power efficient, and a fast one which you can switch off entirely, might be a viable strategy. You may find that the 32768 Hz oscillator for an RTC can be used as your "slow clock" for undemanding applications, and those are typically optimised for low power consumption. Otherwise, a 1MHz clock might be a good companion for an 8MHz or 12MHz "fast clock".

[backspace119]

Chromatix wrote:

My advice would be to avoid the complication of a continuously-variable clock source, and instead consider switching cleanly between certain chosen discrete clock sources. These can themselves, for flexibility, be integer divisions of a fast master clock; use a presettable counter and a JK flipflop wired to toggle every time the counter reaches zero (and reloads). I think there is a circuit for glitch-free clock source switching somewhere, or just stuff a new value into the preset register.

This was my original thought, but when I saw variable oscillators on mouser I started looking into them to see if they might be a viable alternative (and allow greater flexibility in the clock)

Chromatix wrote:

Usually, slow devices are accessed in a minority of cycles, so it makes performance sense to implement wait states to access them; all the other cycles are then still fast. Power saving, meanwhile, can be achieved through "race to sleep", assuming your static power consumption is low - that is, execute WAI and make your software interrupt driven.

Interrupt driven software is definitely the direction I'll be moving when programming this, but as for the clock, it's less about power consumption and more about being able to get as many cycles as possible without becoming unstable.

Chromatix wrote:

Crystal oscillators take a while to start up and stabilise, and might themselves dominate your power consumption in some cases. In that case, having a slow oscillator that might be more power efficient, and a fast one which you can switch off entirely, might be a viable strategy. You may find that the 32768 Hz oscillator for an RTC can be used as your "slow clock" for undemanding applications, and those are typically optimised for low power consumption. Otherwise, a 1MHz clock might be a good companion for an 8MHz or 12MHz "fast clock".

Using the clock from the RTC is not a bad idea, I need to read up on your suggestions though and figure out the inner workings of switching between discrete clocks. (or switching dividers of a fast clock)

[GARTHWILSON]

backspace119 wrote:

Ya, that's actually another thing, I can only fit two traces through the DIP pins if I'm lucky (on one side of the board that is). I tink it's because my settings are 6mil/6mil, and I think the pads that I've selected for the chips are kind of fat. I assume with SMT style I'd still be able to fit one, so nbd, but I'm not sure if I'd put vias on the pads, since if I had a lot of SMT stuff I might go ahead and build a solder paste dispenser, which works kind of like the big machines do, and it might make a mess.

With .007" trace & space, you can get three traces between pads of a DIP. Use .030" holes and .050" pads.

With .006" trace & space and .045x.055" pads, you can fit four in:

4tracesThruDIP.gif (8.68 KiB) Viewed 718 times

About switching clock frequencies on the fly: 20+ years ago I did a cycle-stretch thing for slow ROM and tested it, and posted about it at viewtopic.php?p=11796#p11796 . Don't use the 4000-series logic I show there. I used that only because I wanted slow-enough logic that if there were any glitches or runt pulses, my slow oscilloscope could catch them. The circuit uses a flip-flop and a quad NOR, which you would want to select from a faster family.

[backspace119]

GARTHWILSON wrote:

With .007" trace & space, you can get three traces between pads of a DIP. Use .030" holes and .050" pads.

Ya, my pads are probably bigger, I'm using the footprint defaults that come with kicad for sockets with long pads (easier to solder)

GARTHWILSON wrote:

With .006" trace & space and .045x.055" pads, you can fit four in:

Since I'm rewiring, I may try and make the pads smaller and do this, my only reservation is soldering with such small pads

GARTHWILSON wrote:

About switching clock frequencies on the fly: 20+ years ago I did a cycle-stretch thing for slow ROM and tested it, and posted about it at viewtopic.php?p=11796#p11796 . Don't use the 4000-series logic I show there. I used that only because I wanted slow-enough logic that if there were any glitches or runt pulses, my slow oscilloscope could catch them. The circuit uses a flip-flop and a quad NOR, which you would want to select from a faster family.

Thanks for this, I've been looking at clock divider ICs and they're all either super slow or very expensive and not quite what I need. I was thinking about building it in discrete logic, which I may have to. I did, however, find one timer IC that looked nice, but it didn't retain any kind of setting you gave it, it just started from whatever you gave it (it was a 4 bit binary counter). I was thinking about using a 4 bit latch (or just 4 flip flops) to hold a state that I wanted for it, and on every pulse of the clock output have it reload that state back into it, if I did this, I'd have to use it like a -divider- EDIT: multiplier, and invert the latch outputs, or I'd end up with the division setting being 0 at startup, which would probably end up with DC output.

[GARTHWILSON]

backspace119 wrote:

Since I'm rewiring, I may try and make the pads smaller and do this, my only reservation is soldering with such small pads

The pins stick out the back of the board the same distance; and since the holes are thru-plated, soldering the pin will get the heat into the pad and solder that too. No problem at all.