def dendrite_time_stepper(time_vec, I_drive, L3, tau_di):
with open('../master_rate_matrix.soen', 'rb') as data_file:
data_array_imported = pickle.load(data_file)
I_di_list__imported = data_array_imported['I_di_list']
I_drive_vec__imported = data_array_imported['I_drive_vec']
master_rate_matrix__imported = data_array_imported['master_rate_matrix']
p = physical_constants()
Phi0 = p['Phi0']
I_fq = Phi0 / L3
I_di_vec = np.zeros([len(time_vec), 1])
for ii in range(len(time_vec) - 1):
dt = time_vec[ii + 1] - time_vec[ii]
if I_drive[ii] > 18.6e-6:
ind1 = (np.abs(np.asarray(I_drive_vec__imported) -
I_drive[ii])).argmin()
ind2 = (
np.abs(np.asarray(I_di_list__imported[ind1]) -
I_di_vec[ii])).argmin()
rate = master_rate_matrix__imported[ind1, ind2]
# linear interpolation
# rate = np.interp(I_drive[ii],I_drive_vec__imported,master_rate_matrix__imported[:,ind2])
else:
rate = 0
I_di_vec[ii + 1] = rate * I_fq * dt + (1 - dt / tau_di) * I_di_vec[ii]
return I_di_vec
def inter_fluxon_interval__fit(I, mu1, mu2, V0):
Ic = 40e-6
R = 4.125
V_fq_vec = (Ic * R) * ((I / Ic)**mu1 - 1)**(mu2) + V0
p = physical_constants()
ifi_vec = p['Phi0'] / V_fq_vec
return ifi_vec
#%%
import numpy as np
from matplotlib import pyplot as plt
from scipy.signal import find_peaks
from scipy.signal import savgol_filter
# from soen_sim import input_signal, synapse, dendrite, neuron
from _plotting import plot_dend_rate_array, plot_dend_time_traces
from _functions import cv, save_session_data, read_wr_data
from util import physical_constants
import pickle
p = physical_constants()
plt.close('all')
#%% set case
num_jjs = 2 # 2 or 4
#%% inputs
# DR loop inductances (in all cases, left side has additional 10pH through which flux is coupled (M = k*sqrt(L1*L2); in this case k = 1, L1 = 200pH, L2 = 10pH))
dL = 1 # pH
L_left_list = np.arange(7, 13 + dL, dL) # pH
L_right_list = np.flip(np.arange(17, 23 + dL, dL)) # pH
num_L = len(L_right_list)
# dendritic firing junction bias current
dI = 2 # uA
def t_fq(I, Ic, R, mu1, mu2):
p = physical_constants()
t_fq_vec = (p['Phi0'] / (Ic * R)) * ((I / Ic)**mu1 - 1)**(-mu2)
return t_fq_vec
def dendritic_time_stepper(time_vec, R, I_drive, I_b, Ic, M_direct, Lm2, Ldr1,
Ldr2, L1, L2, L3, tau_di, mu_1, mu_2):
p = physical_constants()
Phi0 = p['Phi0']
prefactor = Ic * R / Phi0
I_fq = Phi0 / L3
#initial approximations
Lj0 = Ljj(Ic, 0)
Iflux = 0
Idr2_prev = ((Lm2 + Ldr1 + Lj0) * I_b[0] +
M_direct * Iflux) / (Lm2 + Ldr1 + Ldr2 + 2 * Lj0 +
(Lm2 + Ldr1 + Lj0) * (Ldr2 + Lj0) / L1)
Idr1_prev = I_b[0] - (1 + (Ldr2 + Lj0) / L1) * Idr2_prev
Ij2_prev = I_b[1]
Ij3_prev = I_b[2]
I_di_vec = np.zeros([len(time_vec), 1])
# Idr1_next, Idr2_next, Ij2_next, Ij3_next, I1, I2, I3 = dendrite_current_splitting(Ic,0,I_b[0],I_b[1],I_b[2],M_direct,Lm2,Ldr1,Ldr2,L1,L2,L3,Idr1_prev,Idr2_prev,Ij2_prev,Ij3_prev)
# print('Idr1 = {}uA, Idr2 = {}uA, Ij2 = {}uA, Ij3 = {}uA'.format(Idr1_next*1e6,Idr2_next*1e6,Ij2_next*1e6,Ij3_next*1e6))
for ii in range(len(time_vec) - 1):
dt = time_vec[ii + 1] - time_vec[ii]
#dendrite_current_splitting(Ic, Iflux, Ib1, Ib2, Ib3, M, Lm2,Ldr1,Ldr2,L1,L2,L3,Idr1_prev,Idr2_prev,Ij2_prev,Ij3_prev)
Idr1_next, Idr2_next, Ij2_next, Ij3_next, I1, I2, I3 = dendrite_current_splitting(
Ic, I_drive[ii + 1], I_b, M_direct, Lm2, Ldr1, Ldr2, L1, L2, L3,
Idr1_prev, Idr2_prev, Ij2_prev, Ij3_prev)
# I_di_sat = I_di_sat_of_I_dr_2(Idr2_next)
Ljdr2 = Ljj(Ic, Idr2_next)
Lj2 = Ljj(Ic, Ij2_next)
Lj3 = Ljj(Ic, Ij3_next)
# Ljdr2 = Ljj(Ic,0)
# Lj2 = Ljj(Ic,0)
# Lj3 = Ljj(Ic,0)
Idr1_prev = Idr1_next
Idr2_prev = Idr2_next
Ij2_prev = Ij2_next #(Lj3/(L2+Lj2))*I_di_vec[ii]
Ij3_prev = Ij3_next
# I_j_df_fluxon_soen = Phi0/(L1+Ldr2+Ljdr2+Lj2)
# I_j_2_fluxon_soen = Phi0/(Lj2+L_pp)
# I_j_3_fluxon_soen = Phi0/(L3+Lj3)
I_loop2_from_di = (Lj3 / (L2 + Lj2)) * I_di_vec[ii]
I_loop1_from_loop2 = (Lj2 / (L1 + Ljdr2 + Ldr2)) * I_loop2_from_di
# print('I_loop2_from_di = {}uA, I_loop1_from_loop2 = {}uA'.format(I_loop2_from_di*1e6,I_loop1_from_loop2*1e6))
Idr2_next -= I_loop1_from_loop2
Ij2_next -= I_loop2_from_di
Ij3_next -= I_di_vec[ii] - I_loop2_from_di
L_ppp = Lj3 * L3 / (Lj3 + L3)
L_pp = L2 + L_ppp
L_p = Lj2 * L_pp / (Lj2 + L_pp)
# print('L_p = {}pH, L_pp = {}pH, L_ppp = {}pH'.format(L_p*1e12,L_pp*1e12,L_ppp*1e12))
large_number = 1e9
I_flux_1 = 6e-6
I_flux_2 = 20e-6
Ij2_next += I_flux_1 # (Phi0/(L1+L_p))*(L_pp)/(Lj2+L_pp)#(Lj3/(L2+Lj2))*I_di_vec[ii]
# print('Ij2_next += {}uA'.format(1e6*(Phi0/(L1+L_p))*(L_pp)/(Lj2+L_pp)))
# Ij3_next += (Phi0/L_pp)*(L3/(L3+Lj3)) + (Phi0/(L1+L_p))*L3/(Lj3+L3) - I_di_vec[ii]
Ij3_next += I_flux_2 # (Phi0/L_pp)*(L3/(L3+Lj3))
# print('Ij3_next += {}uA'.format(1e6*(Phi0/L_pp)*(L3/(L3+Lj3))))
# print('term_1 = {}; term_2 = {}'.format( (Phi0/L_pp)*(L3/(L3+Lj3)) , (Phi0/(L1+L_p))*L3/(Lj3+L3) ) )
if Idr2_next > Ic:
factor_1 = inter_fluxon_interval(
Idr2_next) # ( (Idr2_next/Ic)**mu_1 - 1 )**(-mu_2)
else:
factor_1 = large_number
if Ij2_next > Ic:
factor_2 = inter_fluxon_interval(
Ij2_next) # ( (Ij2_next/Ic)**mu_1 - 1 )**(-mu_2)
else:
factor_2 = large_number
if Ij3_next > Ic:
factor_3 = inter_fluxon_interval(
Ij3_next) # ( (Ij3_next/Ic)**mu_1 - 1 )**(-mu_2)
else:
factor_3 = large_number
# print('factor_1 = {}, factor_2 = {}, factor_3 = {}'.format(factor_1,factor_2,factor_3))
r_tot = (factor_1 + factor_2 + factor_3)**(-1)
I_di_vec[ii + 1] = r_tot * I_fq * dt + (1 - dt / tau_di) * I_di_vec[ii]
return I_di_vec
def run_sim(self):
sim_params = self.synapse_model_params
tf = sim_params['tf']
dt = sim_params['dt']
time_vec = np.arange(0, tf + dt, dt)
p = physical_constants()
# input_spike_times = self.input_spike_times
#setup input signal
if hasattr(self, 'input_signal'):
self.input_spike_times = self.input_signal.spike_times
else:
self.input_spike_times = []
#here currents are in uA. they are converted to A before passing back
I_sy = self.synaptic_bias_current * 1e6
#these values obtained by fitting to spice simulations
#see matlab scripts in a4/calculations/nC/phenomenological_modeling...
if 'gamma1' in sim_params:
gamma1 = sim_params['gamma1']
else:
gamma1 = 0.9
if 'gamma2' in sim_params:
gamma2 = sim_params['gamma2']
else:
gamma2 = 0.158
if 'gamma3' in sim_params:
gamma3 = sim_params['gamma3']
else:
gamma3 = 3 / 4
#these fits were obtained by comparing to spice simulations
#see matlab scripts in a4/calculations/nC/phenomenological_modeling...
# tau_rise = (1.294*I_sy-43.01)*1e-9
tau_rise = (0.038359 * I_sy**2 - 0.778850 * I_sy - 0.441682) * 1e-9
self.integration_loop_total_inductance = self.integration_loop_self_inductance + self.integration_loop_output_inductance
_reference_inductance = 775e-9 #inductance at which I_0 fit was performed
_scale_factor = _reference_inductance / self.integration_loop_total_inductance
#I_0 = (0.06989*I_sy**2-3.948*I_sy+53.73)*_scale_factor #from earlier model assuming 10uA spd
# I_0 = (0.006024*I_sy**2-0.202821*I_sy+1.555543)*_scale_factor # in terms of currents
I_0 = 1e6 * (2.257804 * I_sy**2 - 76.01606 * I_sy + 583.005805) * p[
'Phi0'] / self.integration_loop_total_inductance # in terms of n_fq
#I_si_sat is actually a function of I_b (loop_current_bias). The fit I_si_sat(I_b) has not yet been performed (20200319)
I_si_sat = 19.7
tau_fall = self.integration_loop_time_constant
# I_si_vec = synaptic_response_function(time_vec,input_spike_times,I_0,I_si_sat,gamma1,gamma2,gamma3,tau_rise,tau_fall)
I_si_vec = np.zeros([len(time_vec), 1])
for ii in range(len(time_vec)):
I_si_vec[ii] = synapse_time_stepper(time_vec, ii,
self.input_spike_times, I_0,
I_si_sat, gamma1, gamma2,
gamma3, tau_rise, tau_fall)
self.I_si = I_si_vec * 1e-6
self.time_vec = time_vec
return self
