Handshaking & Pipelining

Generally speaking (in this context), handshaking is the process of electronic systems agreeing on what to do next. In NCL, this takes the form of modules requesting a NULL or DATA signal.

The request for NULL indicates that the operation has been performed, and the results saved for the next module to use.
The request for DATA indicates that the module has reset, passed the reset signal to the next module, and is ready for more work.

If you’re having trouble, think of it like two people doing laundry, one is operating the washing machine, the other is operating the dryer. Additionally, assume that the operators can’t be sure of how long a load will take in either stage. It’s not a perfect analogy, but it might help. Initially, both are empty:

In the empty state, the washing machine operator knows he or she needs clothes, so he or she finds some and puts them in the wash. The drier is also empty, but it needs to get its clothes from the washer, so it has to wait.
When the washer is finished, the operator takes out the clothes and pus them between the washer and dryer, for the dryer’s operator to use when ready. Since the washer is now ‘reset’, it is ready for clothes again, its operator loads it.
The dryer operator sees the clothes are ready to be dried and puts them in the dryer. Eventually they finish and leave the system.

Now, imagine that after a few loads, the dryer can’t keep up, and the washing machine finishes a load while the dryer is still running. The load goes into the space between the washer and dryer. Since the washer can’t know how close the dryer is to completion, it has to wait for the dryer to finish and take the waiting load before starting another.

July 10, 2017

Null Convention Logic Gates

See this post for information about NCL signals.

Symbol

It may or may not surprise you to know that NCL has it’s own set of gates for building circuits. They are called threshold gates and are drawn as

with the ‘threshold number’ in the middle. The inputs come in on the left (round) part, and the output comes out the point on the right.

Operation

For all threshold gates:

If all inputs are 0, the output is 0
If the number of inputs that are set is >= the threshold number, the output is 1
Otherwise, it holds its value.

threshold gates can have input weights, meaning that an input can count towards the threshold count more than once.

Notation

A threshold gate is denoted as THmn where m is the threshold value (the number of inputs needed to transition to a 1) and n is the total number of inputs. If weights are needed, then THmnW## is used, where the #’s are replaced with the weights (order is irrelevant). A TH23W2 gate sets when the first input is set (as it is counted twice), or when both of the other 2 are set. The boolean logic function is

TH23W2 Set: A+BC

Note that when just B or C are set, the gate keeps its value, so if it was a 1, then it will stay a 1, and if it was a 0 it will stay a 0. The logic function for clearing the gates is essentially the same for all:

THm1 Clear: A'
THm2 Clear: A'B'
THm3 Clear: A'B'C'
THm4 Clear: A'B'C'D

Summary

Threshold gates are used in NCL because they are less volatile than standard gates. Once they have an output, they hold on to int until their inputs are completely cleared (circuit reset). This means that a NCL component’s output will not flip more than once in an operation. This property of threshold gates means that if the outputs are set, the component has completed, and if they are reset, it has cleared. You always know when it is ready to move on.

July 9, 2017

Null Convention Logic Signals

If you don’t yet know what asynchronous logic is, I recommend reading this post (or using Google).

For the time being, I will be focusing my efforts on Null Convention Logic (NCL). NCL is technically a separate logic system from boolean logic. For now, I’ll stick to considering it in the context of a boolean system. What that means is that all signals in the design will still be boolean, they will just represent something a little different.

In NCL, each signal (boolean value) still has 2 meanings, but they are changed. A value of 0 means NO_DATA/NULL and a value of 1 means DATA. Different signals have different Data assigned. To represent a standard boolean variable, 2 lines are needed, the encoding:

FALSE (0): DATA0=1, DATA1=0
TRUE (1): DATA0=0, DATA1=1
Null/Unknown: DATA0=0, DATA1=0

The new state, NULL is used when the circuit hasn’t determined a value for the line yet. Detecting completion of the circuit is then as simple as seeing if one line in each group is set. When all the outputs are assigned a value, the values are saved, and the circuit resets to NULL outputs.

Note, it is entirely possible to have more than two lines in a group. Imagine If I made a circuit that took in an 8-bit color value and outputted if it was mostly one of {Red, Green, Blue}. Instead of having 2 boolean variables for 3 states (wasting a combination), I can have 1 three-state variable: {DATA_RED, DATA_BLUE, DATA_GREEN}. When the circuit is reset, the output is unknown, so all three go to NULL, when the inputs are received and the result found, the circuit sets one of the lines. This is not that different from ‘one-hot’ encoding in synchronous logic, except that it does have a defined all-off state that carries meaning (not data, but it does indicate circuit state). In synchronous logic, such a state would be an error.

The most common encodings I’ve come across so far are 2-rail {DATA0 and DATA1} and 4-rail {DATA0, DATA1, DATA2, DATA3}, representing 1 and 2 boolean variables respectively.

Anyways, that’s the basics of NCL signaling. The thing to remember is that the signals have to go to NULL between calculations, otherwise, there’s no way to know when they are ready.