Implementing a neuromorphic computing system using nanophotonic neurons

1. A photonic neuron, comprising:

an excitatory-input photo detector comprising a first photodiode that converts an optical excitatory input signal into a corresponding electrical excitatory input signal;

an inhibitory-input photo detector that converts an optical inhibitory input signal into a corresponding electrical inhibitory input signal;

an electrical neuron that receives the electrical excitatory and inhibitory input signals, and generates an electrical output signal, which includes periodic voltage spikes that are triggered by integration of the electrical excitatory and inhibitory input signals;

a light-emitting output device comprising a second photodiode, which converts the electrical output signal into a corresponding optical output signal;

an integrating capacitor coupled between an internal node and ground;

a first transistor having a gate coupled to the internal node;

a second transistor having a gate coupled to a source of the first transistor and having a source coupled to ground;

a first resistor coupled between the internal node and a drain of the first transistor; and

a second resistor coupled between the gate of the second transistor and ground;

wherein the light-emitting output device comprises a light-emitting diode (LED) or a laser.

2. The photonic neuron of claim 1 , wherein the electrical neuron implements an integrate-and-fire model, wherein the electrical excitatory and inhibitory input signals are integrated until a firing threshold is reached, which causes the electrical neuron to fire and generate a voltage spike on the electrical output signal.

3. The photonic neuron of claim 2 , wherein the electrical neuron implements a sigmoid-shaped nonlinear activation function.

4. The photonic neuron of claim 1 ,

wherein the electrical excitatory input signal comprises a weighted sum of one or more excitatory input signals; and

wherein the electrical inhibitory input signal comprises a weighted sum of one or more inhibitory input signals.

5. A neuromorphic computing system, comprising:

a sequence of neural layers, wherein each neural layer comprises a set of photonic neurons, wherein each photonic neuron converts one or more optical input signals into corresponding electrical input signals, processes the electrical input signal to produce an electrical output signal, and converts the electrical output signal into an optical output signal; and

a set of intervening optical interconnection networks, which couple together successive neural layers in the sequence of neural layers, wherein each interconnection network provides weighted links between photonic neurons in successive neural layers;

wherein each photonic neuron comprises:

a first photodiode coupled between an internal node and a drain voltage v d ;

a second photodiode coupled between ground and the internal node;

an integrating capacitor coupled between the internal node and ground;

a first transistor having a gate coupled to the internal node;

a light-emitting output device coupled between a drain of the first transistor and the drain voltage v d ;

a second transistor having a gate coupled to a source of the first transistor and having a source coupled to ground;

a first resistor coupled between the internal node and a drain of the first transistor; and

a second resistor coupled between the gate of the second transistor and ground.

6. The neuromorphic computing system of claim 5 , wherein each optical interconnection network in the set of intervening optical interconnection networks comprises a network of 2×2 Mach-Zehnder interferometer blocks connected in a mesh.

7. The neuromorphic computing system of claim 5 , wherein:

the first photodiode converts an optical excitatory input signal into a corresponding electrical excitatory input signal;

the second photodiode converts an optical inhibitory input signal into a corresponding electrical inhibitory input signal;

each photonic neuron further comprises an electrical neuron that receives the electrical excitatory and inhibitory input signals, and generates an electrical output signal that includes periodic voltage spikes that are triggered by integration of the electrical excitatory and inhibitory input signals; and

the light-emitting output device converts the electrical output signal into a corresponding optical output signal.

8. The neuromorphic computing system of claim 7 , wherein the electrical neuron implements an integrate-and-fire model, wherein the electrical excitatory and inhibitory input signals are integrated until a firing threshold is reached, which causes the electrical neuron to fire and generate a voltage spike on the electrical output signal.

9. The neuromorphic computing system of claim 7 , wherein the electrical neuron implements a sigmoid-shaped nonlinear activation function.

10. The neuromorphic computing system of claim 7 ,

wherein the light-emitting output device comprises a light-emitting diode (LED) or a laser.

11. The neuromorphic computing system of claim 7 ,

wherein the electrical excitatory input signal comprises a weighted sum of one or more excitatory input signals; and

wherein the electrical inhibitory input signal comprises a weighted sum of one or more inhibitory input signals.

12. A three-dimensional (3D) integrated neuromorphic computing system, comprising:

a stack containing computational planes, wherein each computational plane in the stack includes a nanophotonic neural network;

wherein the stack additionally includes at least one memory plane containing nanophotonic memory elements; and

wherein successive planes in the stack are coupled together through a set of vertical optical vias, each of which comprises:

a bottom waveguide configured to convey light in a horizontal direction;

an inverted 45-degree reflector configured to reflect horizontally propagating light from the bottom waveguide in a vertical direction;

a vertical silicon escalator configured to convey light in a vertical direction from the inverted 45-degree reflector;

a 45-degree reflector configured to reflect vertically propagating light from the vertical silicon escalator in a horizontal direction; and

a top waveguide configured to convey light received from the 45-degree reflector in a horizontal direction.

13. The 3D integrated neuromorphic computing system of claim 12 , wherein each nanophotonic neural network in each computational plane comprises:

a sequence of neural layers, wherein each neural layer comprises a set of photonic neurons, wherein each photonic neuron converts one or more optical input signals into corresponding electrical input signals, processes the electrical input signal to produce an electrical output signal, and converts the electrical output signal into an optical output signal; and

a set of intervening optical interconnection networks, which couple together successive neural layers in the sequence of neural layers, wherein each interconnection network provides weighted links between photonic neurons in successive neural layers.

14. The 3D integrated neuromorphic computing system of claim 13 , wherein each optical interconnection network in the set of intervening optical interconnection networks comprises a network of 2×2 Mach-Zehnder interferometer blocks connected in a mesh.

15. The 3D integrated neuromorphic computing system of claim 13 , wherein each photonic neuron in each neural layer comprises:

an excitatory-input photo detector that converts an optical excitatory input signal into a corresponding electrical excitatory input signal;

an inhibitory-input photo detector that converts an optical inhibitory input signal into a corresponding electrical inhibitory input signal;

an electrical neuron that receives the electrical excitatory and inhibitory input signals, and generates an electrical output signal, which includes periodic voltage spikes that are triggered by integration of the electrical excitatory and inhibitory input signals; and

a light-emitting output device, which converts the electrical output signal into a corresponding optical output signal.

16. The 3D integrated neuromorphic computing system of claim 15 , wherein the electrical neuron implements an integrate-and-fire model, wherein the electrical excitatory and inhibitory input signals are integrated until a firing threshold is reached, which causes the electrical neuron to fire and generate a voltage spike on the electrical output signal.

17. The 3D integrated neuromorphic computing system of claim 15 , wherein the electrical neuron implements a sigmoid-shaped nonlinear activation function.