Sunteți pe pagina 1din 4

Design Problem 1 Documentation

by: Estebat, Chelsea T.


2011-19998
I. Introduction
From the DP specs, I've created a Moore Machine that follows the requirements of the
circuit: A 4 by 4 bit multiplier.
Below will be the story of how the machine was created and how the problem was
solved. :)
II. Data Path
The data path was the easier of the two so I opted to start here. I used an 8-bit register
plus one D-flipflop to store the 9 bit data needed. There will also be a need for A LOT of
multiplexers, a 4 by 4 bit adder and some logic gates but we'll get there.
From the specs, the computation seems to go like this: Load, Check and Add, Shift,
Check and Add, Shift, Check
and Add, Shift, Check and
Add and one last Shift. This
means there's a fixed number
of clock cycles until the
completion (later we'll get
there) and every clock cycle,
we have to do either Check
and Add, or Shift.
I chose to do both in
one clock cycle since it's
easier (or maybe not?) and
cheaper in real life. From the
Specs, it also seems like the
division of clock cycles are
like this: (LD, Chk & Add),
(Shift, Chk & Add), (Shift, Chk
& Add), (Shift, Chk & Add),
(Shift). I opted not to use this
pattern and instead used this:
(LD), (Chk & Add, Shift), (Chk
& Add, Shift), (Chk & Add,
Shift), (Chk & Add, Shift). The
difference between two
patterns is shown on the
figure.
As you can see, both logic arrive at the same answer, although this does seem to need
one more clock cycle and memory element, which is why there are nine bits of storage in the
circuit as will be shown.

From there, the Register


and the Adder were placed like
this.
As shown, the 4 MSBs
enter the adder before being
sent back to the register for a
shift. The Carry out of the
Adder is stored in the extra Dflipflop for the next clock cycle.

Based on the logic shown in the previous


figure, this should work.
The Adder also adds the 4 MSBs of the 9 bit Register to either ProductA or 0000. This
will be dealt with by a 2-1 multiplexer.
Since the value to be added is based on the D0 (the R0 in the next clock cycle) and
whether it is not in State4, the selector used is D0Q4.

III. Control Path


I put the state machine part in control path because as mentioned, it's a Moore
machine, therefore outputs rely hugely on the states. This is the reason why I considered the
states as part of the control.
I mentioned that there would be a fixed number of clock cycles for this machine. This in

turn translates to having a fixed number of states in my machine. From the data path, it
should be obvious that I need 5 states. Aside from having a fixed number of states, it also has
a fixed path, which means I would simply be using a counter from 0 to 4. The only problem
would be the Hold when Start = 0 and count is in State0.
This is what I implemented to hold in State0 if Start = 0.

As for the other states, they are pretty straight forward. Once it gets to State1, it would
just go to the next state after every positive edge of the clock.

IV. Mixing Things Up


Final part would be connecting the Control and the Data path. The data path's behavior
is dependent on the controls, in particular, which data should the data path compute?
This is the purpose of the many multiplexers in the circuit.
First, the input of the 8-bit Register and the 9th bit D-flipflop varies as shown.

Since the selector of the mux is Start and S0, as shown, LD will only happen on the
first computational clock cycle. This effectively initializes the Register and the flipflop in
preparation for (based on the logic mentioned previously) adding, checking and shifting. For
D4 to D7, the reason why ProductA is ANDed to D0 is because from the logic, it will only Load
ProductA if D0 is 1. If this is not done, ProductA will always be loaded to D4 to D7 regardless
if D0 is not asserted, resulting to a failure of computation.
Shift will happen whenever it is not in State0 regardless of the input Start, which means
in the middle of computation, it will always finish no matter the input.
Hold happens when the machine is in State1 but the Start = 0. It just feeds the current
outputs back to the Register and D-flipflop.

V. Notes, Problems, and Things That Aren't Really Important But I Put Them Anyway
Because This DP Messed Me Up (And There's Another Half Page to Fill Up)
Now that I reread this Documentation for editing, I realized there's something wrong
with the Logic Used Figure. The first clock cycle doesn't ADD ProductA to the Upper Product,
it chooses between LOADING it or not based on the Check. I could have fixed that but doing
that would be too troublesome so I'm just noting it here :D
Also, that was a problem that I encountered. First time I finished the circuit, I was so
happy it worked when I tried it but then I tried 15 times 15 (1111 x 1111 = 11100001) IT GOT
THE WRONG ANSWER. What's worse is that it gets a DIFFERENT WRONG ANSWER if I
don't reset the whole machine, which is a big no-no so I had to trace A LOT of wires just to
find what went wrong.
In conclusion, debugging circuits is as difficultif not moreas debugging a written
program. I have learned my lesson, sensei.
If you noticed that there was more detail on the LD for the mux part (the one on top of
the page) it's because the morning after I finished the circuit, I woke up to the realization that I
forgot to consider that. It was a problem that I personally encountered so no one can blame
me for putting that into detail in the Documentation.
On a final note, I hope I won't get minus points for putting this unrelated/useless/
unnecessary heading to my Documentation. If it will get me deductions, I'll happily print and
submit a version without it :D

S-ar putea să vă placă și