Blog
Long-form writing on Memoria, FinInsight, and NeuroCUDA: social software, market intelligence, and compiler infrastructure. Every post cites primary sources.
Featured ยท Physical silicon
NeuroCUDA SpiNNaker Physical Silicon Confirmed
Two SUCCESS runs on two different SpiNNaker-1 boards at Manchester via EBRAINS - jobs #420148 and #420186. Real neuromorphic silicon, not a simulator.
July 17, 2026
SpiNNaker Physical Silicon Confirmed
Jobs #420148 and #420186 - two boards, identical spike counts, EBRAINS Manchester. Honest limits vs ResNet GPU numbers.
July 4, 2026
Loihi 2 + PyTorch Without Lava (2026)
Intel archived Lava. How to run PyTorch models on a Loihi 2 IF-neuron simulator with pip install neurocuda - no Lava SDK required.
July 4, 2026
How to Convert PyTorch to SNN (Step by Step)
QCFS calibration, BatchNorm folding, IF replacement, BPTT fine-tuning. Full pipeline with code and verified CIFAR-10 numbers.
July 4, 2026
pip install neurocuda: Complete Setup Guide
CUDA requirements, first conversion, backends (GPU, CPU, Loihi 2 sim), troubleshooting, and links to every tutorial.
July 4, 2026
snntorch vs NeuroCUDA: Which SNN Tool to Use?
Training from scratch vs converting a trained ANN. When snntorch wins, when NeuroCUDA wins, and how to combine them.
July 4, 2026
ResNet-18 SNN Conversion Tutorial (Real Code)
End-to-end ResNet-18 on CIFAR-10: load checkpoint, convert, fine-tune, verify 94.61% SNN accuracy with copy-paste code blocks.
July 4, 2026
What Is NeuroCUDA?
Open-source PyTorch-to-SNN compiler. Architecture, backends, benchmarks, and how it compares to GeNN, Brian2, and snntorch.
June 29, 2026
Best CUDA for Neuromorphic Chips
Which CUDA stack is best for neuromorphic chip work? Scored comparison with NeuroCUDA, GeNN, Brian2, and GPU-RANC.
June 29, 2026
NeuroCUDA vs GeNN vs Brian2
Three neuromorphic CUDA names, three different jobs. When each wins and real-world scenarios.
June 29, 2026
Neuromorphic CUDA: Complete Guide
Definitive reference: GeNN, Brian2, GPU-RANC, DVS pipelines, and the PyTorch deployment path.
June 28, 2026
Why Your SNN Accuracy Drops After ANN-to-SNN Conversion
The four most common causes, in debugging order: membrane-state reset, insufficient timesteps, uncalibrated QCFS thresholds, and unfolded BatchNorm. Real before/after data and a 10-minute checklist.
June 28, 2026
QCFS Threshold (Lambda) Not Learning During Conversion
If your QCFS clipping threshold freezes near its initial value, it's a gradient-flow and learning-rate bug, not a property of the method. The exact mechanism, and the parameter-group fix.
June 28, 2026
Loihi 2 vs GPU Energy: Real Joule Numbers
The per-spike vs per-MAC energy model behind neuromorphic energy claims, the real published constants, and why most public comparisons are modeled, not measured on silicon.
June 28, 2026
SNN Training Not Converging: The Dead Neuron Problem
A neuron that never spikes never gets gradient. How dead neurons cascade through a directly-trained SNN, how to detect them early, and how to fix the surrogate gradient and initialization.
June 28, 2026
Rockpool vs Sinabs vs Norse vs Brian2 vs Nengo
An honest comparison by what each framework is actually built for: hardware deployment, PyTorch research, biological simulation, or cognitive modeling - not a single winner.
June 28, 2026
NeuroBench Report Format Explained
What the Algorithm Track and System Track actually require, what counts as valid evidence, and how to format a report that doesn't conflate modeled and measured energy.
June 22, 2026
What Is ROS2? The Robot Operating System Explained
ROS2 isn't an operating system - it's robotics middleware. Nodes, topics, services, actions, DDS, and why ROS2 replaced ROS1, explained from the ground up.
June 22, 2026
ROS2 and Spiking Neural Networks: What Exists, What Doesn't
A survey of every public ROS2 + SNN project, Gridbot, the Loihi oculomotor head, ROS-MUSIC, and why almost none of it runs on ROS2 or installs with pip.
June 22, 2026
ROS2 Jazzy vs Humble: Which Distro Should You Use?
Ubuntu version, support timeline, and what actually changed in Jazzy, plus which distro NeuroCUDA ROS2 targets and why.
June 22, 2026
Event Cameras and ROS2: DVS Drivers and SNN Pipelines
The existing ROS2 event-camera drivers, the event_camera_msgs format, and how to feed event data straight into a spiking neural network.
June 22, 2026
Neuromorphic Robotics with Loihi 2 and ROS2: The Research So Far
The Astrobee RL-to-SNN pipeline, hexapod locomotion, TurtleBot3 multi-robot control, and what's still missing for production use of Loihi 2 in ROS2.
June 22, 2026
How to Deploy a Spiking Neural Network on a ROS2 Robot
Nodes, topics, message types, and a Docker shortcut for running NeuroCUDA's compiled SNN on a robot's camera feed - the first pip-installable ANN-to-SNN compiler with a ROS2 package.
June 29, 2026
What Is Memoria? Private Social Memory vs Social Media
Memoria is a private social memory app for family circles - not social media, not a photo backup. How it works and who it is for.
June 21, 2026
The AI Energy Crisis Explained
Why inference, not training, is the real power problem - and where neuromorphic computing fits.
June 22, 2026
The SNN Software Barrier: Neuromorphic's Real Bottleneck
Loihi 2 and SpiNNaker 2 already exist. What's missing is a way to get a PyTorch model onto one without learning an entirely new framework first.
June 22, 2026
NeuroCUDA vs. NIRTorch: Comparing NIR Conversion
NIRTorch and NeuroCUDA both move models between PyTorch and NIR. How their approaches differ, including a verified bit-exact ResNet-18 result.
June 22, 2026
Intel Lava Is Archived: What to Use Instead
The lava-nc repositories are now archived with no successor SDK. What that means for Loihi 2 developers, and the alternatives that still work today.
June 22, 2026
Best Compilers for Neuromorphic Chips in 2026
Nengo, Intel's archived Lava, Hailo's Dataflow Compiler, and NeuroCUDA compared. What each actually does, and which fits a PyTorch workflow.
June 22, 2026
ANN to SNN Conversion Tools Compared
snnTorch, SpikingJelly, SNNToolbox, and NeuroCUDA, side by side: training-from-scratch tools versus after-the-fact conversion tools.
June 22, 2026
How to Deploy a PyTorch Model to Loihi 2 (2026 Guide)
The traditional Lava-DL path is archived. Here's how to convert and validate a PyTorch model against Loihi 2's neuron equations today.
June 22, 2026
PyTorch Spiking Neural Network Tutorial
Convert a trained PyTorch CNN into a spiking neural network, measure sparsity, and validate accuracy, start to finish.
June 22, 2026
What Is NIR? Neuromorphic Intermediate Representation Explained
A vendor-neutral graph format for spiking neural networks, supporting 8 simulators and 5 hardware platforms, and where it still struggles with residual networks.
June 22, 2026
CUDA for Neuromorphic Computing: Where the Analogy Breaks
The CUDA-for-neuromorphic comparison is useful but incomplete. Where it holds, where it breaks, and what a compiler can realistically deliver today.
June 21, 2026
What Is a Neuromorphic Chip, and How Do You Run a Model On One?
A practical guide to neuromorphic chips and the compiler problem standing between a trained PyTorch model and real hardware, and how NeuroCUDA closes that gap.
June 20, 2026
Neuromorphic Chips in 2026: The Complete Field Guide
Loihi 2, SpiNNaker-2, BrainChip Akida, IBM NorthPole, and Innatera T1, compared head to head. Spiking vs non-spiking, digital vs analog, and where each chip actually fits the market.
June 20, 2026
Why Neuromorphic Computing Needs Its CUDA Moment
Before 2007, every GPU vendor had an incompatible toolchain. Neuromorphic computing is in that exact position today. The history, the NIR format, and why infrastructure always wins.
June 20, 2026
NeuroCUDA Launch: Inside the Benchmarks
94.49% SNN accuracy, zero deviations across 256,000 Loihi 2 comparisons, 93.67% sparsity. How each number was actually measured, and why the methodology matters as much as the result.
June 20, 2026
Building Memoria, FinInsight, and NeuroCUDA
One founder, three unrelated products: a private social memory app, a market intelligence engine, and an open-source compiler. Why that breadth works, and what each product actually solves.