Intel Loihi

Intel Loihi is the most advanced research neuromorphic processor available. The Lava software framework enables programming SNNs for Loihi and simulating them on conventional hardware.

Loihi 2 Architecture

Loihi 2 is Intel's second-generation neuromorphic research chip, featuring significant improvements over the original:

• 128 neurocores per chip, each with up to 8,192 neurons
• Programmable neuron models: Custom neuron dynamics via microcode, not just LIF
• Graded spikes: Spikes can carry integer payloads, not just binary events
• 3-factor learning rules: On-chip learning with reward signals for reinforcement learning
• Enhanced connectivity: Faster spike routing and larger fan-out per neuron
• Scalability: Multiple chips connected via chip-to-chip interfaces (Pohoiki Springs: 768 Loihi chips)

The Lava Framework

# Install Lava
pip install lava-nc

# Basic Lava process example
from lava.proc.lif.process import LIF
from lava.proc.dense.process import Dense
from lava.proc.io.source import RingBuffer as Source

# Create a simple SNN
input_layer = Source(data=spike_data)
weights = Dense(weights=weight_matrix)
neurons = LIF(shape=(n_neurons,), vth=10, du=1, dv=1)

# Connect layers
input_layer.s_out.connect(weights.s_in)
weights.a_out.connect(neurons.a_in)

Lava Programming Model

1. Define Processes

   Processes are the basic computational units in Lava. Each process has ports (inputs/outputs) and internal state variables.

2. Connect Processes

   Connect output ports to input ports to define the network topology. Data flows as spikes between processes.

3. Choose a Backend

   Run on CPU (PyProcess), GPU (CuPyProcess), or Loihi hardware (NxProcess). Same code, different backends.

4. Execute

   Run the network for a specified number of timesteps. Lava handles synchronization and data flow.

Training for Loihi

There are two main approaches to getting trained models onto Loihi:

Accessing Loihi

• Intel Neuromorphic Research Community (INRC): Apply for cloud access to Loihi systems through Intel's research program.
• Lava Simulator: Develop and test on your local machine using CPU/GPU backends. No Loihi hardware required.
• Kapoho Bay / Pohoiki Springs: USB and rack-mounted Loihi systems available through INRC partnerships.