8bit CPU project had problems - ***SOLVED***

[MK14]

Ignoring the breadboard size/construction, for now ...

I only had a quick glance at your schematic, and was trying to hint that it would seem to have a number of potential weaknesses to it.
Sorry!

Examples are things like...

You may be trying to use 74S logic, which probably is best done on a proper PCB, with really decent layouts and probably ground planes. Because the 74S logic series are meant for very high speed/performance. Which needs more care in its constructions. Because it can easily achieve 100 MHz and very fast changing signals, compared to slower 74LS logic and many other slower types.
Very fast transitioning logic, is especially problematic, on weak noisy breadboards, with significant stray capacitance (inductance etc).

The 74S series needs LOTS of current to drive it, especially at the low logic level. The lower power consumption 74LS is NOT very capable of this (ignoring buffer versions). You can just about get away with driving one 74S connection with LS, and sometimes 2 (maybe more, depending on what inputs and 74 devices you are talking about).
I have NOT gone through your circuit and thoroughly checked how many things are connected to what, but a quick glance worries me.

You seem to have a large number of unconnected inputs. These are best properly sorted out (there are guides to help you, it varies depending on which 74 series it is), as they can easily cause incorrect/odd behavior if left like that.
I.e. They can randomly toggle between 1 and 0, tending to be 1, for most of the time, usually.

tl;dr
Although you are currently running it at 1 and 2 Hz, the glitches etc (because of unconnected inputs, missing decoupling capacitors, unsuitable very high speed logic, etc etc) can toggle/change the state of flip-flops, latches, Ram etc etc, and hence bite you in the foot.
I.e. It might only take a few nano seconds or more, for a glitch to corrupt some kind of latch/Ram etc, even though you are clocking it at only 1 or 2 Hz.

-------------------------------------------------------------------------------------------------------------------------------------

Just in case others want to disagree with me here.
I agree that you can probably take one 74S device and wire it up on a breadboard and it will probably work ok. But probably at a significantly lower frequency than if it was on a properly designed PCB.
I also agree that a 74LS might cope with more than 1 or 2 74S inputs, in a simple circuit. Even if the datasheets disagree.

But to try and make a massive/complicated computer with a huge number of 74S devices and many 74LS (and other series) ones, on a huge number of breadboards. Is not going to be the happiest of experiences as regards it working 100% reliably all of the time.

[MK14]

I found this guy, who seems to be doing similar things to yourself (OP).

He has a fair amount of detail about "glitches" and things like that.

https://leo1cpu.wordpress.com/tag/ttl/

He goes into much detail in places (than I've had the patience/time to explain).

E.g. (very approximate summery) That brief (call it 5 or 10 nano second) "mistakes" in the address decoding while the logic ripples through with the right answer. Can be long enough for the Ram to get corrupted (at the WRONG address), and how he FIXED such problems.

[amyk]

I suspect one of the ICs has its power supply completely disconnected and is getting powered not even through the ESD diodes on its inputs but pure inductive pickup from ambient electromagnetic fields; the amount of current is tiny, but enough to accumulate sufficient charge to work at extremely slow speeds, yet not slightly faster. The time scales you're seeing are roughly on this order of magnitude.

Ironically, the large capacitance of breadboards may actually be helping you with that.

[PauloConstantino]

You mentioned that the clock didn't go very many places.  Is it safe to assume that all the registers are clocked?  If you are trying to use R-S flops, you may be in for a ride.  There's a reason that FPGAs implement synchronous logic and highly discourage asynchronous logic.

The SR latch is only used as a manual clock generator. All my logic is synchronous

[PauloConstantino]

If you want to increase the clock the 1st thing you need to do is to cut that spider web of jumpers you have.

Buy a standard PCB like one of those


Using wirewrap wire mount using the shortest possible connections

A copper plate ground plane will also help.

A second path is to make the same circuit on an FPGA using the same schematics, this way you get really faster clocks

I have 8 high quality gold plated wirewrap boards here waiting to be used. I just wanted the prototype to work before I go to the permanent construction, but it seems the prototype will have to be cancelled

[PauloConstantino]

I suspect one of the ICs has its power supply completely disconnected and is getting powered not even through the ESD diodes on its inputs but pure inductive pickup from ambient electromagnetic fields; the amount of current is tiny, but enough to accumulate sufficient charge to work at extremely slow speeds, yet not slightly faster. The time scales you're seeing are roughly on this order of magnitude.

Ironically, the large capacitance of breadboards may actually be helping you with that.

I have checked all power pins on every IC, they're all receiving power... still the problem is there.


I constructed another circuit, which was a monochrome composite video generator, that reads a ROM and displays an image pixel by pixel. There are many jumper wires in it and its built on breadboards. It works fine, except that some of the pixels *sometimes* do not display, but if I pull the jumpers apart the glitches disappear.

[PauloConstantino]

Let me see if I can describe the problem more accurately.

The CPU is executing a JMP instruction. There is a ROM and a RAM. ROM is from 0x00 to 0x7F. RAM is from 0x80 to 0xFF. The CPU starts at 0x00, and there is a JMP to address 0x80. At 0x80 there is a JMP back to 0x00. The cpu was supposed to just loop there.

It starts well and executes the first JMP. It has to do 3 steps (3 clock cycles) to do this.  It does it very well, with all control bits shown correctly via LED's. (there are 32 control bits).

Then when it gets to 0x80, it loads the JMP, and jumps back to 0x00. It does this a few times, and then it glitches, and the control bits are junk.


Since it works a few times, why would it glitch afterwards? Very strange. The fast I go on the clock the faster it glitches.

Here's the microcode for the FETCH routine.


0x00: 00100000, 00100001, 10000001, 00001000, * PC_Out, IMMVal=0, IMMOut, ALU=1001, ALUMode=0, CarryInMux=01, MAR_Wrt, type=seq(00) *
     00001000, 00000000, 00000010, 00000000, * RD, IR_Wrt, type=seq (00)*
     00100011, 10100001, 00000001, 00001100; * PC_Out, PC_Wrt, ALU=1001, ALUMode=0, CarryInMux=01, IMM_Out, IMMVal=1, type=post-fetch(11) *



and here's the microcode for the JMP instruction.

0x1B: 00100000, 00100001, 10000001, 00001000, * PC_Out, MAR_Wrt, IMM_Out, IMM(0), CarryInMux1=0, CarryInMux0=1, ALUOp=1001, ALUMode=0, Next=00 *
   00001000, 00000000, 01000100, 00000000, * RD, MDRIn_Mux=1, MDR_Wrt, next=00 *
   00100010, 10100001, 00001000, 00001000; * MDR_X_Out, PC_Wrt, IMM_Out, IMM(0), ALUOp=1001, ALUMode=0, CarryInMux1=0, CarryInMux0=1, type=10(pre_fetch) *


Each line is a clock cycle's worth. Each column is one byte from each of the 4 ROMS.

[PauloConstantino]

I'm not using 74S or 74LS. As I said I am using 74HC. The schematics don't show the actual chips I'm using, they are just what was available in the eagle library. They were made for myself as a sketch to help build the CPU.

The schematics have no weakness to it, as you're trying to attack my work again.

The unused inputs in the schematics do not imply my actual circuit have unused inputs either. It doesn't.

The problem is not one with the design as you're trying to suggest. The problem is something else.

[rs20]

You mentioned that the clock didn't go very many places.  Is it safe to assume that all the registers are clocked?  If you are trying to use R-S flops, you may be in for a ride.  There's a reason that FPGAs implement synchronous logic and highly discourage asynchronous logic.

The SR latch is only used as a manual clock generator. All my logic is synchronous

We never really seemed to explore this in detail. What inputs are you giving to the SR latch? Do you have two pushbuttons, one going to R, the other going to S, and alternating between them to generate your clock? This is a strange, but I suppose reasonable, way of debouncing your clock input.

Failing that, just about any other solution will require debouncing, even if an SR latch is somewhere in the circuit (garbage in to SR flip flop --> garbage out).

[PauloConstantino]

You mentioned that the clock didn't go very many places.  Is it safe to assume that all the registers are clocked?  If you are trying to use R-S flops, you may be in for a ride.  There's a reason that FPGAs implement synchronous logic and highly discourage asynchronous logic.

The SR latch is only used as a manual clock generator. All my logic is synchronous

We never really seemed to explore this in detail. What inputs are you giving to the SR latch? Do you have two pushbuttons, one going to R, the other going to S, and alternating between them to generate your clock? This is a strange, but I suppose reasonable, way of debouncing your clock input.

Failing that, just about any other solution will require debouncing, even if an SR latch is somewhere in the circuit (garbage in to SR flip flop --> garbage out).

The SR latch has its 2 inputs pulled up by resistors. I just ground one of the inputs alternately to generate the pulses.

[MK14]

I'm not using 74S or 74LS. As I said I am using 74HC. The schematics don't show the actual chips I'm using, they are just what was available in the eagle library. They were made for myself as a sketch to help build the CPU.

The schematics have no weakness to it, as you're trying to attack my work again.

The unused inputs in the schematics do not imply my actual circuit have unused inputs either. It doesn't.

So you're now saying the schematic and what you have built, are significantly different!

How was I to know that you have done it significantly different to the schematic ?

You are also saying your setups are completely wonky and unreliable:

It works fine, except that some of the pixels *sometimes* do not display, but if I pull the jumpers apart the glitches disappear.

Anyway...

74HC logic, makes a lot more sense, if you are experiencing problems going from 1 Hz to 2 Hz, because the CMOS, if there are any open connections or partial/intermittent connections, can do things like that, as the Mosfets gates (inputs) charge up and discharge, from their very high impedances.

I think you are going to need a huge amount of luck to find the one or more, possible open connections, out of the thousands, in that huge birds nest of yours.

[PauloConstantino]

I'm not using 74S or 74LS. As I said I am using 74HC. The schematics don't show the actual chips I'm using, they are just what was available in the eagle library. They were made for myself as a sketch to help build the CPU.

The schematics have no weakness to it, as you're trying to attack my work again.

The unused inputs in the schematics do not imply my actual circuit have unused inputs either. It doesn't.

So you're now saying the schematic and what you have built, are significantly different!

How was I to know that you have done it significantly different to the schematic ?

You are also saying your setups are completely wonky and unreliable:

It works fine, except that some of the pixels *sometimes* do not display, but if I pull the jumpers apart the glitches disappear.

Anyway...

74HC logic, makes a lot more sense, if you are experiencing problems going from 1 Hz to 2 Hz, because the CMOS, if there are any open connections or partial/intermittent connections, can do things like that, as the Mosfets charge up and discharge, from their very high impedances.

I think you are going to need a huge amount of luck to find the one or more, possible open connections, out of the thousands, in that huge birds nest of yours.

Are you crazy or something? I didn't say they are "Significantly" different. They are the same. Please go bother someone else's posts.

[MK14]

Are you crazy or something? I didn't say they are "Significantly" different. They are the same. Please go bother someone else's posts.

Yes, I'm crazy.
I'm so crazy, that I thought that a 74S is different to a 74HC.
I'm even crazier for thinking that a schematic which shows lots of unconnected inputs, is different to one with everything connected.

[MK14]

Since it works a few times, why would it glitch afterwards? Very strange. The fast I go on the clock the faster it glitches.

I did not realize you had used lots of 74HC logic. I'm pleased you have, as it is probably better overall for something like this.

Now that I realize that you have, that is EXACTLY the sort of symptom, I would expect to see with CMOS logic devices. Because any open circuits (quite possible with your breadboard construction), would just be floating about in the wind. Fluctuating between 1 and 0, somewhat randomly. Even moving your hand nearby, could switch it on and off, over a period of seconds.
CMOS logic is quite well known to do this.

There are other fault possibilities as well. But one or more, possibly intermittent open circuits, can act like you describe.

tl;dr
CMOS inputs, if open circuit, can randomly change state every second or so (time period can be almost anything). Similar to what you have been experiencing.

[PauloConstantino]

Ok sorry if I was rude.

I have grounded all of the unused inputs... Even if they are not part of the circuit....

[Ian.M]

Both Siwastaja and myself are *VERY* suspicious of the quality of those jumpers.  Unless you've got a receipt for them from Sparkfun or other well respected 'Maker' supplier, or from a reputable major distributor,  the odds are they are probably made with wire with a very low CSA and strand count, and there is a significant chance the wire may not even be copper. 

Cheap Chinese jumpers are often made with CCA or CCS wire or recycled low grade copper - more like a poor grade of brass - and tend to have much higher resistance than a similar copper wire should have.  CCA and CCS are probably the worst as anything that breaks through the thin flash of copper plate over the Aluminum or mild steel core + even traces of moisture (not uncommon during shipping) results in rapid bimetallic corrosion, especially at the crimped terminals.   All of them fatigue far more readily than copper wire and are prone to strand breakage

If you do want to keep using pre-made 'dupont' terminated stranded jumpers, they all need testing on a milliohmmeter before use, to eliminate bad ones and its also worth gently pulling on them end to end to try to eliminate 'stretchy' ones that are full of broken strands.  A sample from each purchase should also be closely examined to check the ends are properly crimped (with no horror storries like the strands being folded back over the insulation then the connector roughly crushed over the lot), and CCA wire should be eliminated by destructive testing (use the flame test - CCA strands go soft and droop at a far lower temperature than copper strands), and CCS wire eliminated by testing if the strands are attracted to a magnet.

However insulated 22AWG solid core tinned copper wire is dirt-cheap (just strip it from scrap Ethernet cable, but not stranded patchleads), easy to work with, doesn't have separate end pieces so eliminates two potential bad connections per wire and doesn't clutter the breadboard so much,  and when its going bad, its very obvious as it becomes floppy at the break.  Also the ease of cutting it to length will eliminate most of the birdsnest.

[MK14]

Ok sorry if I was rude.

I have grounded all of the unused inputs... Even if they are not part of the circuit....

Good, that should improve the reliability.
With CMOS, it is especially important to ground (or logic 1) the inputs (even if unused). Because if they float towards mid-supply voltage they can cause the IC to misbehave and/or use too much current.

I like 74HC logic over older 74 TTL, because they can use considerably less power and have big fan in/out figures. They are also much more available than the outgoing older 74 TTL devices, which are getting increasingly harder to get hold of. They also have nice big output voltage swings, which makes them potentially less susceptible to noise.
Great for battery powered applications as well.

[PauloConstantino]

Both Siwastaja and myself are *VERY* suspicious of the quality of those jumpers.  Unless you've got a receipt for them from Sparkfun or other well respected 'Maker' supplier, or from a reputable major distributor,  the odds are they are probably made with wire with a very low CSA and strand count, and there is a significant chance the wire may not even be copper. 

Cheap Chinese jumpers are often made with CCA or CCS wire or recycled low grade copper - more like a poor grade of brass - and tend to have much higher resistance than a similar copper wire should have.  CCA and CCS are probably the worst as anything that breaks through the thin flash of copper plate over the Aluminum or mild steel core + even traces of moisture (not uncommon during shipping) results in rapid bimetallic corrosion, especially at the crimped terminals.   All of them fatigue far more readily than copper wire and are prone to strand breakage

If you do want to keep using pre-made 'dupont' terminated stranded jumpers, they all need testing on a milliohmmeter before use, to eliminate bad ones and its also worth gently pulling on them end to end to try to eliminate 'stretchy' ones that are full of broken strands.  A sample from each purchase should also be closely examined to check the ends are properly crimped (with no horror storries like the strands being folded back over the insulation then the connector roughly crushed over the lot), and CCA wire should be eliminated by destructive testing (use the flame test - CCA strands go soft and droop at a far lower temperature than copper strands), and CCS wire eliminated by testing if the strands are attracted to a magnet.

However insulated 22AWG solid core tinned copper wire is dirt-cheap (just strip it from scrap Ethernet cable, but not stranded patchleads), easy to work with, doesn't have separate end pieces so eliminates two potential bad connections per wire and doesn't clutter the breadboard so much,  and when its going bad, its very obvious as it becomes floppy at the break.  Also the ease of cutting it to length will eliminate most of the birdsnest.

I got them from JPR. Here in the UK. They are dupont. Dunno if made in china. So I get from you that I should replace all of them with solid copper wires? In this case, I might just reconstruct everything using wirewrap boards with gold plated pins. Oh well.

[PauloConstantino]

Both Siwastaja and myself are *VERY* suspicious of the quality of those jumpers.  Unless you've got a receipt for them from Sparkfun or other well respected 'Maker' supplier, or from a reputable major distributor,  the odds are they are probably made with wire with a very low CSA and strand count, and there is a significant chance the wire may not even be copper. 

Cheap Chinese jumpers are often made with CCA or CCS wire or recycled low grade copper - more like a poor grade of brass - and tend to have much higher resistance than a similar copper wire should have.  CCA and CCS are probably the worst as anything that breaks through the thin flash of copper plate over the Aluminum or mild steel core + even traces of moisture (not uncommon during shipping) results in rapid bimetallic corrosion, especially at the crimped terminals.   All of them fatigue far more readily than copper wire and are prone to strand breakage

If you do want to keep using pre-made 'dupont' terminated stranded jumpers, they all need testing on a milliohmmeter before use, to eliminate bad ones and its also worth gently pulling on them end to end to try to eliminate 'stretchy' ones that are full of broken strands.  A sample from each purchase should also be closely examined to check the ends are properly crimped (with no horror storries like the strands being folded back over the insulation then the connector roughly crushed over the lot), and CCA wire should be eliminated by destructive testing (use the flame test - CCA strands go soft and droop at a far lower temperature than copper strands), and CCS wire eliminated by testing if the strands are attracted to a magnet.

However insulated 22AWG solid core tinned copper wire is dirt-cheap (just strip it from scrap Ethernet cable, but not stranded patchleads), easy to work with, doesn't have separate end pieces so eliminates two potential bad connections per wire and doesn't clutter the breadboard so much,  and when its going bad, its very obvious as it becomes floppy at the break.  Also the ease of cutting it to length will eliminate most of the birdsnest.

0.7 omh each

[MK14]

0.7 omh each

Although I totally agree with the advise you are being given, about possible duff or poor quality "wires". I would like to add that many (especially cheaper, ebay and hence often Chinese sourced/made) breadboards, also have bad reputations for having very bad connections.

[tautech]

Ok sorry if I was rude.

I have grounded all of the unused inputs... Even if they are not part of the circuit....

Good, that should improve the reliability.
With CMOS, it is especially important to ground (or logic 1) the inputs (even if unused). Because if they float towards mid-supply voltage they can cause the IC to misbehave and/or use too much current.

I like 74HC logic over older 74 TTL, because they can use considerably less power and have big fan in/out figures. They are also much more available than the outgoing older 74 TTL devices, which are getting increasingly harder to get hold of. They also have nice big output voltage swings, which makes them potentially less susceptible to noise.
Great for battery powered applications as well.

This ^.
RTFM datasheet. Don't tie them high or low is the standard recipe for disaster.
Does nobody read datasheets these days or do they think they know better that the manufacturer. 

[Ian.M]

0.7 Ohms each is absolutely horrendous. For comparison 24AWG electrical grade copper wire has a resistance of about 0.085 Ohms per meter, and 22AWG about 0.053 Ohms per meter.  As your jumpers are well under a meter long they apparently have two orders of magnitude too much resistance - I suspect that either they are total shite or your meter doesn't read low ohms very reliably.

Have you got some that aren't in use, or can you pull a reasonable sample without loosing track of what you need to reconnect?

As I said, test them with a low-ohm meter, and compare with the same length of 22AWG tinned copper wire.  You'll soon spot the bad ones.   Most hand-geld DMMS are fairly useless for trying to measure resistances under one Ohm. If you have an ESR meter for testing caps, its Ohms scale *may* go low enough.

Otherwise, if you don't have a low-ohm meter - Build one!

Take a LM317 regulator and connect a 15 Ohm resistor with a 100 Ohm preset in parallel, adjusted to 12.5 Ohms total, between Out and Adj to make a 100mA current source,  make up a x10 gain differential amplifier using a 5V single supply rail-to-rail OPAMP and 1% or better resistors, and make a pair of DIY Kelvin clips from clothespegs and self-adhesive copper foil.

Use a stable 5V supply to power the 100mA current source, connect the current source output (Adj pin) and ground to the forcing contacts, and the x10 amplifier to the sense connections.  Use a DMM on mV for the readout, directly in milliohms. 

If you have a precision one Ohm resistor put it in the Kelvin clips and calibrate it by adjusting the preset for a reading of 1.000V.  It will read up to 2 ohms from a 5.0V supply.   Don't use it for in-circuit testing of anything containing small signal semiconductors ar ICs as the internal junctions wont stand up to the forcing current.

You can also make up psuedo-Kelvin probes for checking breadboard contact resistance - take some gold plated pins or offcuts of gold plated component lead thin enough not to damage the breadboard contacts, and solder two wires side by side to each, taking care not to let solder run down the pin.  The protruding pin length should be just enough to fully seat in the breadboard.  Insulate and protect the joint with small-bore heatshrink sleeving, then use one wire to each pin for the forcing current and the other for the sense voltage. That will measure the series resistance of two breadboard-pin contacts and the strip between them + the resistance of the pin between the wires and the contact point which should be negligible.

[MK14]

0.7 omh each

Please don't take offense at this question.

But did you initially short together the meter probes and see what it reads ?
Because some of that 0.7 Ohms could be the meter + probes resistance.

[Muxr]

You're probably dealing with multiple issues, one of which is likely chatter, like the signal is being affected by voltage droops and all that stray capacitance and inductance. As others have said, solderless breadboards and jumpers don't really scale well for big projects. Things become unmanageable real quick. They are really meant for smallish quick prototypes.

Anyways, a scope can probably tell you more than we can.

[PauloConstantino]

0.7 omh each

Please don't take offense at this question.

But did you initially short together the meter probes and see what it reads ?
Because some of that 0.7 Ohms could be the meter + probes resistance.

Yes buddy. I think my multimeter just isnt very precise.