chaogate

Simulations of the 3-MOSFET system known as a choagate using pyspice.

This library corresponds to a selection of the code used in "Chaogate Parameter Optimization using Bayesian Optimization and Genetic Algorithm", “AI for Hardware Security", ISQED’21, 4/7/2020.

Install

pip install chaogate

How to use

import chaogate
from chaogate import *

Ensure you change the global_path to the folder containing the as described in the core page.

By default, they point to:

global_path

'C:\\Anaconda3\\Lib\\site-packages\\PySpice\\Examples\\libraries\\chaogate'

We can instantiate a chaogate with the chaogate function, though this is automatically called by all the other functions which need it in this library. Most functions, such as sweep, grid, and bifurcate, accept kwargs as an argument, which refers to all the possible keyword arguments of the choagate netlist. For exmaple,

s1,s2=sweep(Vbias=(0,1.2,0.01),TEMP=(-20,120,20))
plot_sweep(s1,ncurves=10)
plot_sweep(s2,ncurves=10)

sweep : Elapsed time: 0.3116 seconds

Once we sweep, we calculate the relevant chaotic properties using iterate.

This is packaged into an automatic process, along with calculating the lyapunov exponent, with the bifurcate function.

We can calculate these quantities over entire regions of parameter space with functions like grid, then estimate the computational stability with boolean_divergence:

g=bifurcate(Vbias=(0,1.2,0.01),Vdd=(1.15,1.25,0.001),as_grid=True)

grid : Elapsed time: 29.5447 seconds
iterate_map : Elapsed time: 5.9907 seconds

plot_bifurcate(itr=g.iterate.sel(Vdd=1.2),lya=g.lyapunov.sel(Vdd=1.2))

grad = boolean_gradient(g.iterate)
div = boolean_divergence(grad)
plt.matshow(div)
plt.colorbar()

<matplotlib.colorbar.Colorbar at 0x255f89bae88>

We can quickly compare the difference between the desired chaotic properties (positive $\lambda$), and the undesirable properties (the boolean divergence) by taking a linear combination such as:

normalized_lyapunov = (g.lyapunov.data>0)*g.lyapunov.data/np.max(g.lyapunov.data)
normalized_divergence = div/np.max(div) 
optimizer = normalized_lyapunov - normalized_divergence 
plt.matshow(optimizer)
plt.colorbar()

<matplotlib.colorbar.Colorbar at 0x255f4b6be48>

With an appropriate optimizatier we can then begin to assess the appropriate chaogate design

sidis.refresh()