Project Details

Since I was born in mid-eighties, I've never experienced real 74xx build. Few gates here and there was common, but lager designs were practically unneeded - as microcontrollers, CPLDs or FPGAs replaced need for such as work.
But my possession in vintage systems and desire to understand inner working principles of modern devices brought me to design of simple CPU made of simple 74xx devices.

CPU design
CPU is not that complicated circuit, if we keep goals simple - so no hardware multipliers, few registers, no fancy addressing modes. Such as CPU is not particular useful, as even old 8080 outperforms this easily, but performance is not primary goal now.
Basically, CPU does simple things - moving data from/to memory locations and making operations on it. Program flow should be allowed to change and IO latches are needed for real world operation.


Registers
Registers are elementary part of CPU design. They serve as most used and useful memory locations, source or target of most of instructions. We will have a few registers in our design.

Heart of register is latch 74HCT574. It latches data from data bus DB on rising clock of WE signal. Passing this data to DB is controlled by OE signal, using bus driver 74HCT245. Theoretically we can use OC signal of 574, but data should be accessible even when output is not brought to DB. That is why two ICs are needed to build single register.
We can have a lot of registers on single bus, with WE and OE signals for each one register.

ALU
Adding ALU is quite simple task, thanks to 74181 developed in late 60's/start of 70's. It is 4-bit wide ALU, capable of performing almost all common logical and arithmetic operations.
Let's put two registers together, add 74181 and serve with single bus driver.

Nothing special here, but this starts to be quite useful. We have two registers (A and B), controlled by respective OE and WE signals, ALU, with operation controlled by signals on M and S1 to S4 signals (for more details see 74181 datasheet). Because 74181 doesn't have tristate outputs for connecting to DB, bus driver is needed here. So, data in both registers (accessible from DB) can be passed through ALU and put on DB again.
Imagine we want to do this sequence: put data to A, another data to B, perform ALU operation and put into A again. We need to put A data on bus, assert and release AWE, then put B data on bus, assert and release BWE. In the meantime, ALU does its job (it is only combinational logic) and on F1 to F4 outputs is result. We can assert ALUOE to put result on bus. To write it to A register, asserting AWE is needed... but wait. If we assert WE, latched data (ALU result) appears on data lines of A register, ALU changes its output and this is (or may be) transferred to A register.
That's why third register is needed. Let’s call it T - temporary register. After putting ALU content on bus, we write it to T register and then (when ALU output is securely saved) to A register again.

PC
Let's focus now on another important part of CPU, program counter - PC. It's main job is to increment whenever new instruction is needed or set to value when program jump is to be made.

Nothing special again. Two chained 74HCT193 counters, EEPROM memory holding program and instruction register (IR). It holds current instruction byte until it is fully executed.
Preset inputs of counters (A, B, C and D) are connected to BD, in order to allow direct change of PC (program jump). Otherwise PC changes after each single instruction by CLOCK UP signal (pin 5).


Instruction decoder, part one
PC and registers with ALU are muscles of CPU, doing hard work, but it needs a brain - to decide when and how to change control signals. Instruction decoder does this job. Now starts the real fun and messing with 74xx logic.
Before actually building instruction decoder, it is necessary to decide which instructions we are going to decode.


Instructions
For this computer, I decided to use only three instructions:
1, load direct data to A
2, move data from source to destination. Source can be A, B, RAM or input registers; destination can be A, B, PC, RAM or output registers.
3, do ALU operation between A and B, move result to A
Allowing PC to be result of move allows jumps. You can transfer input data from IO port to RAM in single instruction. From hardware point of view, RAM is treated as another register, with address bus connected to B register. So, B is address pointer for RAM operation. Some move instructions have to affect on registers or memory. Example is move A to A. This could be equivalent of NOP instruction.
There is no dedicated indirect addressing register, no stack, no interrupts.
MSB of instruction determines whether instruction is LDI. We need to waste only one bit for this, so 7 bits are used as immediate data. As immediate data are one of sources for jump instructions, this allows addressing 128B of program ROM. In fact, data from ALU (computed jump) can be used for jumping, but this address is only 4 bits wide, allowing addressing 16B of ROM, leaving this option as not very useful.
If MSB is zero, next bit determines MOV or ALU instruction - notice how this step by step description determines real operation of instruction decoder.

Instruction timing
Instructions are divided into single steps. In our case, we will have for steps, let's call it machine cycles.
M1: load instruction to IR and put source data on DB
M2: load source data from bus to T register
M3: put data from T register on DB
M4: load data from DB to destination, increment PC

Black rectangles denote active (high) level. CLK is incoming clock signal. Whole instruction is done in eight cycles.

Instruction set is simple:

If actual instruction is MOVI, source data is lower 7 bits from IR, destination is A
If actual instruction is MOV, source data is determined by IR[3..5] and destination by IR[0..2]
If actual instruction is ALU, source data is from ALU bus driver, destination is A
This gives us first clue about instruction register operation.
Notice leading edge of M2 comes while M1 is still high. This overlap is needed to securily write data into T register. The same goes for M3 and M4.
Building clock circuit is quite simple. We need D-flip-flop, dividing input signal by two, giving with incoming clock four possible states. Those states are decoded by simple AND logic. To achieve 1:1 duty cycle of incoming clock signal from 555 timer, second D-FF is used.

Instruction decoder, part two
Knowing what a how to decode, we can proceed in design of instruction decoder. Let's start with most complicated instruction, MOV. We need to select source register during phase M1 and put on bus - so OE signal of selected register should be active during M1 phase. We can use 74HCT138 1-of-8 decoder. Fortunately it has three chip select pins, two of them inverted. We can connect those two to IR[7] and IR[6] signals, thus activating during MOV instruction. Third, high active, select pin is connected to M1 signal. The same goes for selecting destination register, with the exception that third chip select pin goes to M4 signal. To complete MOV instruction, we need to take care of T register. OE of T register will be active during M3 and WE during M2. MOVI and ALU instructions are very alike, except of that first one select IROE signal, while former selects ALUOE signal during M1. AWE (write to A register) is active during M4 for both instructions.

IC20, IC21 and IC22 does this job - generates IROE and ALUOE signals, as well as AWE signal. For this purpose I used simple looking, but useful software, Logic Friday.
I generated this truth table for AWE signal

and software minimized this table into equations and generated circuit of logic gates doing the same job.

I did the same for IROE and ALUOE signals. Voila, instruction decoder is done.
We need to make jumps conditional in some way. I decied to use register B for this purpose. When it's content is 0xF, jump (MOV to PC) is executed as NOP.

Notice, on final schematics, signal M3 is not used at all. It is needed for latching output of T register, but M1 is used, as driver expects negative logic and M3 is only inverted M1.

Input/output ports
The only thing not described for now is IO part. We have two signals from 138 decoders, so all is needed is double 4-bit bus driver (IC25) for input ports and two 4-bit wide latches as output ports (IC26, IC27).

Programming
As our CPU is basically complete, we need to program it to make something useful. Lets start with simple program - emulation of four NAND gates. 

MOV IA,A ; move data from input A to register A
MOV IB,B ; move data from input B to register B
ALU NAND ; do NAND operation
MOV A,PA ; move data from A (ALU result) to port A
MOVI 0 ; move zero to A
MOV A,B ; move this zero to B
MOV A, PC ; jump to zero

Quick hand assembly gives this output 0x20

0x29
0x54
0x04
0x80
0x01
0x02

That is ready to be burned into EEPROM. I used Genius G540 programmer - really low cost, but it does its job.

Result, or 7400^2 to 7400^x

Circuit was built on perfboard with dimension cca 18x18cm. Current consumption is about 180mA, majority of this is drawn by 74181 and 74175 in plain old TTL technology.
Clock speed is determined by C1 capacitor. For 1uF, clock generator ticks at about 80Hz, giving 10Hz execution speed. For no capacitor, oscillator works at frequency given by stray capacitance, resulting in approx 57kHz execution speed. Yes, whopping 57,000 instructions per second.

Processor, or single board computer, works as expected. I wrote emulation program that allows emulation of four NAND gates, basically acting like single 7400 IC - let's call it second generation 7400. This may seem to be trivial and unusable (OK, it IS unusable), but limited number of those (second generation) 7400 ICs allows to build another CPU that allows emulation of another 7400 - third generation 7400. We can continue indefinitely, building more and more generations of 7400 ICs. If we look at last generation of 7400, we can zoom at its basic parts - there would be 7400 computers, built from 7400 computers - something like zooming on fractals. Fractal 7400 computer, that's it.