Comparator construction
In the upcoming lab, you will need to generate a signal that behaves according
to the following CSA statement:
h_synch <= '1' when ((h_count >= 100) and (h_synch < 200)) else '0';
However, if you implemented the h_synch signal using this method (non-gated
output), your circuit would not behave correctly. The fundamental problem
is that
non-gated signals generate glitches on their output. These
glitches are generated by the comparators inherent in the CSA above. Fully
understanding this problem and its solution will require some discussion, so let's
get started.
Large comparators are realized by performing a comparison from the most to
least significant bit. At each step, a bit of X is compared to a bit of Y.
If the bits are equal, then continue to the next least significant bit.
Otherwise, either X or Y is larger, and the comparison is over. In this way,
the validity of the output ripples from the most to the least significant bit.
Since real logic gates have delay, there is a delay to get a valid bit. This
delay is the focus of this lecture.
Gated and non-gated circuits
In the upcoming lab, you will have to design a circuit to output 2 signals crucial to the
proper operation of a VGA display. In order for these outputs to properly drive the
display, they need to be both glitch free and promptly available after a clock edge. This
requires that the outputs be gated - they must be output from flip flops, not combinational
logic. In order to demonstrate the potential pitfalls of combinational logic, I coded-up
lec05.vhdl (linked at the top of this page).
The outputs from the two processing blocks depend on the value of the counter's output
(processQ) compared to one of three constants: L0, L1, L2 (set inside lec05.vhdl). Both
the gated and non-gate circuit have the same relationship between processQ and their
output JB - the difference is how this output is generated. The table below illustrates.
Combinational Realization Non-gated | Sequential Realization Gated |
JB(3 downto 0) <= "0001" when ((processQ >= 0) and (processQ < L0)) else
"0010" when ((processQ >= L0) and (processQ < L1)) else
"0100" when ((processQ >= L1) and (processQ < L2)) else
"1000";
|
process(clk)
begin
if (rising_edge(clk)) then
if (reset = '0') then
JB(6 downto 4) <= "000";
elsif ((processQ >= 0) and (processQ < L0)) then
JB(6 downto 4) <= "001";
elsif ((processQ >= L0) and (processQ < L1)) then
JB(6 downto 4) <= "010";
elsif ((processQ >= L1) and (processQ < L2)) then
JB(6 downto 4) <= "100";
elsif (processQ >= L2) then
JB(6 downto 4) <= "111";
end if;
end if;
end process;
|
At first glance, you would not be faulted for thinking that these two circuits have
identical outputs. Theoretically, they do, but reality has a cruel surprise in the
the form of glitches. If you connect an oscilloscope up to the PMOD connector (see
page 20 of the Nexys Video Board Reference Manual) corresponding
to pins JB(3) and JB(2) (the most 2 significant bits of the non-gated
comparator outputs), you would see the trace below.
You should notice that the output is not entirely smooth in the area between the edges.
These discontinuities illustrate the fact that the comparator is built from bit-slice
comparators and, as a consequence, the output is generated by rippling a signal from
the most significant bit of the comparator to the least. This allows the output to
temporarily assume an incorrect output until the rippling output reaches the least
significant bit. While the discontinuities look small, this is only the portion of
the signal that has reached the output pin of the FPGA. Internally, the glitches
have a much larger magnitude.
Buttons
You should notice that the counter only counts with btn(4) = '0'. A quick glance
at the silk screen on the Nexys Video board will confirm that this is the center button
in the 4-way pad on the lower right of the board. The buttons are nominally logic
0, meaning that when they are not pressed, they are at logic 0. When pressed the
buttons go to logic 1.