TRIMUX in silicon (regular CMOS process)

[Shaos]

I found the way to build TRIMUX on regular CMOS technology (it could be ordered for silicon production on existing factories):



Here you can see one ternary multiplexer/demultiplexer (half of TRIMUX).

P.S. It's appeared that I reinvented this circuit that was published as part of Master thesis in one Russain university 15 years ago...

[Shaos]

This is similar schematics built from my reusable CMOS-blocks (with NAND/NOR as ternary-binary converter):



Where GREEN is ternary path (actually analog), BLACK is binary path, BLUE is negative voltage and RED is positive voltage...

Below is test results of it in LTspice IV:

[Shaos]

Another way to do this:



Input inverters are assembled from transistor of different sizes to shift thresholds higher and lower in order to get 4 binary signals (left image below) which are used to get another 6 binary signals to control 3 analog CMOS-switches without overlapping (right image below) to avoid short circuits on switches' inputs:

---

P.S. It's similar to http://www.scribd.com/doc/78370674/Tern ... chitecture (pp.139-148) but purely on CMOS without diodes or resistors and with short circuit protection as described above.

[Shaos]

It is possible to achieve the same results without transistor modification (using standard CMOS-blocks built from identical transistors):



Voltage curves are very similar to what we had above:

[Shaos]

This is complete ternary selector (multiplexor-demultiplexor) with gaps to avoid short-circuits during transitions (see above):

[Shaos]

And this is ternary selector with gaps (to have "break-before-make" non-shorting transition) in silicon packaged into SOIC16 package



P.S. This is a piece of the silicon die responsible for ternary selector:



P.P.S. Testing (update March 2016):





Resulting oscillograms (connected as buffer and inverter respectively):

AnalogDiscovery2-mux3buf.png

AnalogDiscovery2-mux3inv.png

[xakepp35]

Молодец. Impressive work!
Possibility of doing chips is "one step closer" to 8086-like stuff.
Which frequency could it achieve?
What is the cost?

More interesting, could you do, for example, SRAM? a 27-trit on-chip register(s)?
That would be much more convenient than doing memory registers from tons of trimuxes.
Also could be convenient for simplier projects like ternary calc, clock, conter, so on where memory is required to store something.

[Shaos]

This one is slow (<20kHz), but next one hopefully will achieve 100 MHz

[xakepp35]

Ok, i think frequency is just a matter of silicon tech process.
So what do you think about SRAM chip?
This is quite basic block of every PC of any architecture.
And there is quite nothing there left to research on a "switch level".

A great challenge is to fit ternary CPU on a motherboard-sized PCB, so that you can fit a computer on several PCBs.
So making bigger blocks as sram register and ternary adder would be definetely a big step forward.
I saw your hackday project, it great, but is too big =)
Is hard to repeat for average hobbyst at home and almost not useful.
Because slow, no memory, no I/O, no OS, you can only run simple algorithms.
Having such chips commercially availble would allow easier life for cpu inventors and interest for software developers.

And its a somewhat real, practical task.
For example: "to develop ternary computer which is capable of serving static web pages", like this guy did.

[Shaos]

Ternary SRAM itself is useless without ternary microprocessor, so I want to make ternary microprocessor first

[xakepp35]

Hello! I was wondering, do you have any success on this project? Especially, about embedding higher trimux count in a single chip?

[Shaos]

Not yet...