8000 GitHub - can-lehmann/verilog2neur: Let's synthesize McCulloch-Pitts Neural Networks from Verilog · GitHub
[go: up one dir, main page]

Skip to content

About

Let's synthesize McCulloch-Pitts Neural Networks from Verilog

Resources

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

9 Commits

Folders and files

Repository files navigation

Synthesizing McCulloch-Pitts Neural Networks from Verilog

McCulloch-Pitts neurons are a simple model of biological neurons. It is possible to build a functionally complete set of logic gates using McCulloch-Pitts neurons. This means that any digital circuit can be represented using McCulloch-Pitts neural networks. verilog2neur is a tool that synthesizes McCulloch-Pitts neural networks from Verilog descriptions of digital circuits. It supports combinational as well as sequential circuits.

Adding 2 + 3 using Neurons

Consider the following Verilog module that implements a 4-bit adder:

module top(input [3:0] a, input [3:0] b, output [3:0] c);
  assign c = a + b;
endmodule

Running verilog2neur on this module produces the following neural network description:

_always/1* : _always/1* .
a_0/1* : n4/1* .
b_0/1* : n4/1* .
a_0/1* : n5/2* .
b_0/1* : n5/2* .
_always/1* : n6/1* .
n5/2* ; n6/1* .

...

n50/1* : n51/2* .
n51/2* : c_3/1* .
_always/1* .

We can now add e.g. the numbers 2 (0010) and 3 (0011) by providing the corresponding input bits to the module.

_always/1* : a_1/1* .
_always/1* : b_0/1* .
_always/1* : b_1/1* .

Simulating this network in the neur simulator produces the following output:

00 _always/1 
01 _always/1 n6/1 n14/1 n18/1 n30/1 n34/1 n46/1 n50/1 a_1/1 b_0/1 b_1/1 
02 _always/1 n6/1 n14/1 n18/1 n30/1 n34/1 n46/1 n50/1 a_1/1 b_0/1 b_1/1 n12/1 n4/1 n13/2 n24/2 
03 _always/1 n6/1 n18/1 n30/1 n34/1 n46/1 n50/1 a_1/1 b_0/1 b_1/1 n12/1 n4/1 n13/2 n24/2 n15/2 n7/2 n25/1 
04 _always/1 n6/1 n18/1 n30/1 n34/1 n46/1 n50/1 a_1/1 b_0/1 b_1/1 n12/1 n4/1 n13/2 n24/2 n7/2 n25/1 n16/1 c_0/1 n32/1 
05 _always/1 n6/1 n18/1 n30/1 n34/1 n46/1 n50/1 a_1/1 b_0/1 b_1/1 n12/1 n4/1 n13/2 n24/2 n7/2 n25/1 c_0/1 n32/1 n19/2 n35/2 
06 _always/1 n6/1 n18/1 n30/1 n34/1 n46/1 n50/1 a_1/1 b_0/1 b_1/1 n12/1 n4/1 n13/2 n24/2 n7/2 n25/1 c_0/1 n32/1 n35/2 c_1/1 c_2/1 
07 _always/1 n6/1 n18/1 n30/1 n34/1 n46/1 n50/1 a_1/1 b_0/1 b_1/1 n12/1 n4/1 n13/2 n24/2 n7/2 n25/1 c_0/1 n32/1 n35/2 c_2/1 
08 _always/1 n6/1 n18/1 n30/1 n34/1 n46/1 n50/1 a_1/1 b_0/1 b_1/1 n12/1 n4/1 n13/2 n24/2 n7/2 n25/1 c_0/1 n32/1 n35/2 c_2/1 

As you can see, it stabilizes with the output bits c_0 and c_2 activated, which corresponds to the binary representation of 5 (0101).

A Counter

Now let's look at a slightly more complex example: a 4-bit counter that increments its value on each clock cycle.

module top(input clock, output [3:0] count);
  reg [3:0] counter = 0;
  
  always @(posedge clock)
    counter <= counter + 1;
  
  assign count = counter;
endmodule

Simulating the synthesized neural network for this module produces the following output over several clock cycles:

00 _always/1 _clock/1 
01 _always/1 n7/1 n12/1 n19/1 n26/1 n9/1 n15/1 n22/1 n29/1 n34/1 
02 _always/1 n7/1 n12/1 n19/1 n26/1 n35/1 
03 _always/1 n7/1 n12/1 n19/1 n26/1 n36/1 
04 _always/1 n7/1 n12/1 n19/1 n26/1 n37/1 
05 _always/1 n7/1 n12/1 n19/1 n26/1 n38/1 
06 _always/1 n7/1 n12/1 n19/1 n26/1 n39/1 
07 _always/1 n7/1 n12/1 n19/1 n26/1 n40/1 
08 _always/1 n7/1 n12/1 n19/1 n26/1 n41/1 
09 _always/1 n7/1 n12/1 n19/1 n26/1 n42/1 
10 _always/1 n7/1 n12/1 n19/1 n26/1 n43/1 
11 _always/1 n7/1 n12/1 n19/1 n26/1 _clock/1 
12 _always/1 n7/1 n12/1 n19/1 n26/1 n8/2 n15/1 n22/1 n29/1 n34/1 
13 _always/1 n7/1 n12/1 n19/1 n26/1 counter_0/1 n35/1 
14 _always/1 n12/1 n19/1 n26/1 counter_0/1 n10/1 count_0/1 n36/1 
15 _always/1 n12/1 n19/1 n26/1 counter_0/1 n10/1 count_0/1 n13/2 n37/1 
16 _always/1 n12/1 n19/1 n26/1 counter_0/1 n10/1 count_0/1 n13/2 n38/1 
17 _always/1 n12/1 n19/1 n26/1 counter_0/1 n10/1 count_0/1 n13/2 n39/1 
18 _always/1 n12/1 n19/1 n26/1 counter_0/1 n10/1 count_0/1 n13/2 n40/1 
19 _always/1 n12/1 n19/1 n26/1 counter_0/1 n10/1 count_0/1 n13/2 n41/1 
20 _always/1 n12/1 n19/1 n26/1 counter_0/1 n10/1 count_0/1 n13/2 n42/1 
21 _always/1 n12/1 n19/1 n26/1 counter_0/1 n10/1 count_0/1 n13/2 n43/1 
22 _always/1 n12/1 n19/1 n26/1 counter_0/1 n10/1 count_0/1 n13/2 _clock/1 
23 _always/1 n12/1 n19/1 n26/1 counter_0/1 n10/1 count_0/1 n13/2 n9/1 n14/2 n22/1 n29/1 n34/1 
24 _always/1 n12/1 n19/1 n26/1 n10/1 count_0/1 n13/2 counter_1/1 n35/1 
25 _always/1 n7/1 n12/1 n19/1 n26/1 n13/2 counter_1/1 n10/1 count_1/1 n36/1 
26 _always/1 n7/1 n12/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 n13/2 n37/1 
27 _always/1 n7/1 n12/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 n13/2 n38/1 
28 _always/1 n7/1 n12/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 n13/2 n39/1 
29 _always/1 n7/1 n12/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 n13/2 n40/1 
30 _always/1 n7/1 n12/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 n13/2 n41/1 
31 _always/1 n7/1 n12/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 n13/2 n42/1 
32 _always/1 n7/1 n12/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 n13/2 n43/1 
33 _always/1 n7/1 n12/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 n13/2 _clock/1 
34 _always/1 n7/1 n12/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 n13/2 n8/2 n14/2 n22/1 n29/1 n34/1 
35 _always/1 n7/1 n12/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 n13/2 counter_0/1 n35/1 
36 _always/1 n12/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 n13/2 counter_0/1 n11/2 n16/2 count_0/1 n36/1 
37 _always/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 n13/2 counter_0/1 n11/2 n16/2 count_0/1 n17/1 n37/1 
38 _always/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 counter_0/1 n11/2 n16/2 count_0/1 n17/1 n20/2 n38/1 
39 _always/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 counter_0/1 n11/2 n16/2 count_0/1 n17/1 n20/2 n39/1 
40 _always/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 counter_0/1 n11/2 n16/2 count_0/1 n17/1 n20/2 n40/1 
41 _always/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 counter_0/1 n11/2 n16/2 count_0/1 n17/1 n20/2 n41/1 
42 _always/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 counter_0/1 n11/2 n16/2 count_0/1 n17/1 n20/2 n42/1 
43 _always/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 counter_0/1 n11/2 n16/2 count_0/1 n17/1 n20/2 n43/1 
44 _always/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 counter_0/1 n11/2 n16/2 count_0/1 n17/1 n20/2 _clock/1 
45 _always/1 n19/1 n26/1 counter_1/1 n10/1 count_1/1 counter_0/1 n11/2 n16/2 count_0/1 n17/1 n20/2 n9/1 n15/1 n21/2 n29/1 n34/1 
46 _always/1 n19/1 n26/1 n10/1 count_1/1 n11/2 n16/2 count_0/1 n17/1 n20/2 counter_2/1 n35/1 
47 _always/1 n7/1 n19/1 n26/1 n17/1 n20/2 counter_2/1 n18/2 n23/2 count_2/1 n36/1 
48 _always/1 n7/1 n12/1 n26/1 n20/2 counter_2/1 n17/1 count_2/1 n24/1 n37/1 
49 _always/1 n7/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n27/2 n38/1 
50 _always/1 n7/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n20/2 n39/1 
51 _always/1 n7/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n20/2 n40/1 
52 _always/1 n7/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n20/2 n41/1 
53 _always/1 n7/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n20/2 n42/1 
54 _always/1 n7/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n20/2 n43/1 
55 _always/1 n7/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n20/2 _clock/1 
56 _always/1 n7/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n20/2 n8/2 n15/1 n21/2 n29/1 n34/1 
57 _always/1 n7/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n20/2 counter_0/1 n35/1 
58 _always/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n20/2 counter_0/1 n10/1 count_0/1 n36/1 
59 _always/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n20/2 counter_0/1 n10/1 count_0/1 n13/2 n37/1 
60 _always/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n20/2 counter_0/1 n10/1 count_0/1 n13/2 n38/1 
61 _always/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n20/2 counter_0/1 n10/1 count_0/1 n13/2 n39/1 
62 _always/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n20/2 counter_0/1 n10/1 count_0/1 n13/2 n40/1 
63 _always/1 n12/1 n19/1 n26/1 counter_2/1 n17/1 count_2/1 n20/2 counter_0/1 n10/1 count_0/1 n13/2 n41/1 

We can see that it starts off at 0 in step 0, then activates the counter_0 neuron in step 13 (indicating a count of 1), then switches to counter_1 in step 24 (indicating a count of 2), and continues incrementing the count to 3, 4 and 5 in subsequent steps.

License

This project is licensed under the Apache License 2.0. The neur simulator found in the neur subdirectory is written by Devine Lu Linvega and licensed under the MIT License.

About

Let's synthesize McCulloch-Pitts Neural Networks from Verilog

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages

0