From the sublime to the ridiculous - 8-byte PROM

[Dr Jefyll]

Code: Select all

I'm not sure if something could be done with the diodes to split into two halves, as it were...

I did make a suggestion in this regard (shown again above), but I'll admit my diagram relies on the viewer's imagination. Broadly speaking, a row of decoders (I drew only one decoder) would run horizontally across most of the width of the board. Above the decoder row, and occupying the same width, would be the connection matrix for one nibble, and below the decoder row would be the connection matrix for the other nibble.

Each of the 16 outputs of each decoder needs to route to a diode-array, which in turn feeds upward to one matrix and downward to the other. That's what the diagram is intended to convey, although I've put the diode in a really silly spot, and I'm sure you can do better. The final result might be fairly "busy," with traces running up and down from the diode arrays. But, as you say, space is available (and the board does have two sides ).

-- Jeff

[barnacle]

The other side is probably the best place for the diodes; that way there's only one via _to_ them and their columns can go up and down on the back, as with the muxer design. Lemme have a thunk.

Thing to bear in mind: with the 16-4 bit converter, each column is attached to one and only one row. As it's a ROM, we don't actually have to follow the layout; many of the muxer pins (and/or '154 pins) are attached to each other and don't need to go to the nibble rows, except for one...

My design rules: it has to fit on 100x160 board, double sided, use nothing smaller than SOIC pin spacing (in general) and use an 0.2mm (8 thou) for tracks and clearances; 0.6mm vias with 0.3mm holes.

Neil

[BigDumbDinosaur]

Dr Jefyll wrote:

The new design looks pretty tidy, Neil. But, as always, you can rely on me for a steady stream of nit-picking and background noise!

Don’t feel picked on, Neil.  Jeff’s done the same to me over the years. 

Quote:

Can I suggest you omit these fills? I mean all fifteen of 'em between the bit lines, and the same in the other section. They make the actual bit lines harder to see, and -- more importantly -- harder to cut...

Aside from what Jeff is saying, there is no benefit to be had with using fills in digital work, except possibly as a radiation mitigation method.  If anything, fills add parasitic capacitance to the assembly, which you certainly don’t want if you expect stable operation at anything faster than a 1970s-era pocket calculator.

[barnacle]

Jeff's idea about using double diodes looks as if it might work, but it's a messy board; even with SOT323 (smaller than SOT23-3) the overall width of a block of 16 bytes is somewhat wider than that possible with a SOIC-24W package. That said, it looks as with a bit of creativity, it might work. Move the pullup resistors up/down and shift everything left a little; there are still three chips and a 20-pin header to get on the board.

And the back:

(A lot of cut and paste on there, just to see how things fit).

Neil

[Dr Jefyll]

Quote:

Move the pullup resistors up/down [...]

Hm, that multitude of pullup resistors might be an argument in favor of dropping the priority encoder chips and instead using diode-based nibble encoders. That'd reduce the pullup count from 32 down to 8.

active-low version.png (2.47 KiB) Viewed 1825 times

And, just a reminder, the preliminary nibble encoder layout you posted here could be shrunk significantly if two-diode arrays were used.

Quote:

with a bit of creativity, it might work.

Probably you're already thinking along these lines, but FWIW the left side of the image below suggests a space-saving pattern that cozies up more closely to its neighbours.

-- Jeff

[barnacle]

Your space saving squish has been considered, but for user clarity I would prefer to keep the same nibbles vertically aligned... though of course it's not actually necessary and indeed one could do point-to-point throughout.

But I think on aesthetic grounds, I think I'd rather have all the parts on one side

Just checked the hundreds of diodes I glued on yesterday; just two in the wrong place so easily fixed.

Neil

[barnacle]

Hmm...

Using SOT23 double diodes, for ease of construction; it would be smaller with SOT323 packages but I suspect we could live with 20mm square for each nibble. The select lines (select low) come in from the left, sel0 at the top and sel15 not attached anywhere.

Meanwhile, the original diode prom is failing miserably. For reasons I haven't yet discovered (and won't get around to investigating for a couple of weeks; I'm on holiday paragliding tomorrow) the high voltages are fine but the lows are anywhere from, um, 3v to 4.7v. Not good.

Immediate suspicion on the pull-ups, which are way too small at 10k - what's the impedance of 64 10k resistors in parallel? D'oh! - and I suspect I may have toasted one or more of the '541s. But that's for later.

Neil

[BigDumbDinosaur]

barnacle wrote:

Meanwhile, the original diode prom is failing miserably...Immediate suspicion on the pull-ups, which are way too small at 10k - what's the impedance of 64 10k resistors in parallel?

That would be ~156 ohms, which on five volts, would draw 32 mA, assuming all points were driven to ground.  That’s hardly taxing on an average power supply—even some AA batteries could handle it for a while. 

[barnacle]

Indeed. But it all goes through a single pin of the grounding/selecting IC. There may be paralleling effects going on too that I haven't thought through...

Neil

[BigDumbDinosaur]

barnacle wrote:

Indeed. But it all goes through a single pin of the grounding/selecting IC.

Yikes!  Even 74AC logic would be overloaded.

Quote:

There may be paralleling effects going on too that I haven't thought through...

Relays, man!  Relays.    Of course, relays would be slower than a snake in a snowstorm. 

[L0uis.m]

Hello 0010 all,

barnacle wrote:

Indeed. But it all goes through a single pin of the grounding/selecting IC.

BigDumbDinosaur wrote:

Yikes!  Even 74AC logic would be overloaded.

I didn't dive in to this problem, but maybe buffering or even some multiplexing could help.

[barnacle]

The problem is this, Louis: eight groups of eight bit lines are selected by eight buffers from the address decoding, but each bit line is pulled high, currently with a 10k resistor. When selected, diodes allow the resistor to be shorted to earth via a '154 decoder, which means that each pin on the '154 is potentially getting up to 64 pull-ups (and on average 32) and is trying to pull them to ground. Which it can't... oops!

Bigger resistors might be sufficient, or I may need a rethink of the design (bigger resistors make things slower).

Neil

[BigDumbDinosaur]

barnacle wrote:

Bigger resistors might be sufficient, or I may need a rethink of the design (bigger resistors make things slower).

Maybe boost the signals using some sort of 74ABT driver, e.g., a 74ABT125 or 74ABT541, either which can sink 48+ mA with ease.

74abt125_quad_buffer_driver_TI.pdf

(1017.2 KiB) Downloaded 58 times

74abt541_octal_buffer_driver_TI.pdf

(711.82 KiB) Downloaded 68 times

[L0uis.m]

OK, an other idea.

I had a similar problem with a diode matrix I designed, and was able to reduce the anmount of diodes to less than half of the original design (thereby also reducing the overall current), just by turning logic on it's head (see tables below).

Instead of placing diodes representing ON state
I placed them representing OFF state, reducing the amount by more than 56%❗

[barnacle]

@BDD, those bicmos chips might do to buffer the selector outputs, but it's going to need a respin on the board (if there's even room).
@Louis, the program is already optimised to reduce the number of diodes present, but it's still almost half of what it might be.

Potentially, each output of the '154 is pulling down 64 diodes and associated resistors; on average it's about half of that, but obviously still too much for it. I can't do any practical experiments for a fortnight, but when I get home again.

I do have an idea as to how a solder-through grid might work, too... like a veroboard on steroids. Don't think it will quite fit on a eurocard, but I can stretch a point.

Neil