memristor
Browse files- Memristor.py +188 -0
Memristor.py
ADDED
@@ -0,0 +1,188 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pandas as pd
|
2 |
+
import numpy as np
|
3 |
+
import matplotlib.pyplot as plt
|
4 |
+
from sklearn import preprocessing
|
5 |
+
|
6 |
+
|
7 |
+
class memristor_models():
|
8 |
+
def __init__(self,Roff,Ron,Rint,Amplitude,freq,time_duration,sample_rate,p,j,model):
|
9 |
+
self.initial_Roff=Roff
|
10 |
+
self.initial_Ron=Ron
|
11 |
+
self.Rint=Rint
|
12 |
+
self.Amplitude=Amplitude
|
13 |
+
self.freq=freq
|
14 |
+
self.time_duration=time_duration
|
15 |
+
self.time_duration=time_duration
|
16 |
+
self.sample_rate=sample_rate
|
17 |
+
self.p=p
|
18 |
+
self.j=j
|
19 |
+
self.model= model
|
20 |
+
|
21 |
+
# window functions
|
22 |
+
def jog(self,x1,p):
|
23 |
+
f_x = 1-(((2*x1)-1)**(2*p))
|
24 |
+
return f_x
|
25 |
+
|
26 |
+
def Prodro(self,x1,p,j):
|
27 |
+
f_x = j*(1-(((x1-0.5)**2)+0.75**p))
|
28 |
+
return f_x
|
29 |
+
|
30 |
+
def biolek(self,x1,p,i):
|
31 |
+
if i<0:
|
32 |
+
i=0
|
33 |
+
else:
|
34 |
+
i=1
|
35 |
+
f_x=1-((x1-i)**p)
|
36 |
+
return f_x
|
37 |
+
|
38 |
+
def zha(self,x1,p,j,i):
|
39 |
+
if i<0:
|
40 |
+
i=0
|
41 |
+
else:
|
42 |
+
i=1
|
43 |
+
f_x=j*((1-(0.25*(x1-i)**2)+0.75)**p)
|
44 |
+
return f_x
|
45 |
+
|
46 |
+
def ideal_model(self):
|
47 |
+
start_time = 0
|
48 |
+
time = np.arange(start_time, self.time_duration, 1/self.sample_rate)
|
49 |
+
sinewave = self.Amplitude * np.sin(2 * np.pi * self.freq * time + 0)
|
50 |
+
v_mem=sinewave
|
51 |
+
D=10*pow(10,-9)
|
52 |
+
uv=10*pow(10,-15)
|
53 |
+
delta_R=self.initial_Roff-self.initial_Ron
|
54 |
+
x=(self.initial_Roff-self.Rint)/delta_R
|
55 |
+
x_t=[]
|
56 |
+
i_mem=[]
|
57 |
+
x_t.append(x)
|
58 |
+
R_mem=[]
|
59 |
+
G=[]
|
60 |
+
f1=[]
|
61 |
+
r_val=(self.initial_Ron*x_t[0])+(self.initial_Roff*(1-x_t[0]))
|
62 |
+
R_mem.append(r_val)
|
63 |
+
k=(uv*self.initial_Ron)/(D**2)
|
64 |
+
i_mem.append(0)
|
65 |
+
for i in range(1,len(v_mem)):
|
66 |
+
i_val=v_mem[i]/R_mem[i-1]
|
67 |
+
i_mem.append(i_val)
|
68 |
+
|
69 |
+
if self.model=='Joglekar':
|
70 |
+
f=self.jog(x_t[i-1],self.p)
|
71 |
+
f1.append(f)
|
72 |
+
if self.model== 'Prodromakis':
|
73 |
+
f=self.Prodro(x_t[i-1],self.p,self.j)
|
74 |
+
f1.append(f)
|
75 |
+
if self.model== 'Biolek':
|
76 |
+
f=self.biolek(x_t[i-1],self.p,i_val)
|
77 |
+
f1.append(f)
|
78 |
+
if self.model== 'Zha':
|
79 |
+
f=self.zha(x_t[i-1],self.p,self.j,i_val)
|
80 |
+
f1.append(f)
|
81 |
+
dx_dt=k*i_mem[i-1]*f
|
82 |
+
dx=dx_dt*(time[i-1]-time[i])
|
83 |
+
x=dx+x_t[i-1]
|
84 |
+
x_t.append(x)
|
85 |
+
r_temp=(self.initial_Ron*x)+(self.initial_Roff*(1-x))
|
86 |
+
if r_temp<self.initial_Ron:
|
87 |
+
r_temp=self.initial_Ron
|
88 |
+
if r_temp>self.initial_Roff:
|
89 |
+
r_temp=self.initial_Roff
|
90 |
+
R_mem.append(r_temp)
|
91 |
+
G.append(1/r_temp)
|
92 |
+
|
93 |
+
self.Roff=max(R_mem)
|
94 |
+
self.Ron=min(R_mem)
|
95 |
+
|
96 |
+
return v_mem,i_mem,G,x_t,time,f1
|
97 |
+
|
98 |
+
def plot(self):
|
99 |
+
v,curr,G,x,t,f=self.ideal_model()
|
100 |
+
plt.plot(v,curr)
|
101 |
+
plt.ylabel('i')
|
102 |
+
plt.xlabel('v')
|
103 |
+
plt.show()
|
104 |
+
|
105 |
+
def neural_weight(self,neural_weight,X_max,X_min):
|
106 |
+
self.neural_weight=np.array(neural_weight)
|
107 |
+
|
108 |
+
new_min = (1/self.Roff)
|
109 |
+
new_max = (1/self.Ron)
|
110 |
+
|
111 |
+
# new_weights = []
|
112 |
+
self.mapped_values = []
|
113 |
+
idx = 0
|
114 |
+
for item in self.neural_weight:
|
115 |
+
self.mapped_values.append([])
|
116 |
+
for x in item:
|
117 |
+
scaled_x = ((np.abs(x) - X_min) / (X_max - X_min)) * (new_max - new_min) + new_min
|
118 |
+
if x<0:
|
119 |
+
scaled_x = scaled_x*-1
|
120 |
+
self.mapped_values[idx].append(scaled_x)
|
121 |
+
idx += 1
|
122 |
+
|
123 |
+
# Iterate over the old weights and biases and compute the new values
|
124 |
+
# for weight in neural_weight:
|
125 |
+
# new_weight = ((abs(weight) - X_min) / (X_max - X_min)) * (new_max - new_min) + new_min
|
126 |
+
# new_weight = [(new_weight[i] * -1) if weight[i]<0 else new_weight[i] for i in range(len(new_weight)) ]
|
127 |
+
# new_weights.append(new_weight)
|
128 |
+
|
129 |
+
# self.mapped_values = new_weights
|
130 |
+
|
131 |
+
def variability(self,partition,variability_percentage_Ron,variability_percentage_Roff):
|
132 |
+
v,curr,G,x,t,f = self.ideal_model()
|
133 |
+
self.partition = partition
|
134 |
+
self.variability_percentage_Ron = variability_percentage_Ron
|
135 |
+
self.variability_percentage_Roff = variability_percentage_Roff
|
136 |
+
|
137 |
+
# partitioning
|
138 |
+
self.l2 = []
|
139 |
+
self.l2.append(1/self.Roff)
|
140 |
+
step = ((1/self.Ron)-(1/self.Roff))/(self.partition-1)
|
141 |
+
for i in range(1,self.partition):
|
142 |
+
(self.l2).append(self.l2[0]+(step*i))
|
143 |
+
|
144 |
+
# adding variability to the list
|
145 |
+
new_Goff = (1/self.Roff)+(((1/self.Roff)*self.variability_percentage_Roff)/100)
|
146 |
+
(self.l2).append(new_Goff)
|
147 |
+
|
148 |
+
new_Gon = (1/self.Ron)+(((1/self.Ron)*self.variability_percentage_Ron)/100)
|
149 |
+
(self.l2).append(new_Gon)
|
150 |
+
|
151 |
+
temp = [val for val in self.l2 if (val<=new_Gon and val>=new_Goff)]
|
152 |
+
self.l2 = temp
|
153 |
+
|
154 |
+
self.l2.sort()
|
155 |
+
|
156 |
+
def new_weights(self):
|
157 |
+
self.new_values = []
|
158 |
+
idx = 0
|
159 |
+
|
160 |
+
def closest(lst, K):
|
161 |
+
return lst[min(range(len(lst)), key = lambda i: abs(lst[i]-K))]
|
162 |
+
|
163 |
+
for values in self.mapped_values:
|
164 |
+
self.new_values.append([])
|
165 |
+
for value in values:
|
166 |
+
close_val = closest(self.l2,abs(value))
|
167 |
+
if value <0:
|
168 |
+
close_val = close_val*-1000
|
169 |
+
else:
|
170 |
+
close_val = close_val*1000
|
171 |
+
self.new_values[idx].append(close_val)
|
172 |
+
idx += 1
|
173 |
+
|
174 |
+
return self.new_values
|
175 |
+
|
176 |
+
def Relative_Error (self):
|
177 |
+
Weights_with_var= self.new_weights()
|
178 |
+
self.variability(self.partition,0,0)
|
179 |
+
Weights_without_var= self.new_weights()
|
180 |
+
|
181 |
+
error=[]
|
182 |
+
for i, j in zip(np.array(Weights_without_var), np.array(Weights_with_var)):
|
183 |
+
l = (np.abs(i-j)/i)
|
184 |
+
error.append(l)
|
185 |
+
|
186 |
+
error = np.array(error)
|
187 |
+
return np.abs(np.sum(error))*100/error.size
|
188 |
+
|