FPGAFly: A 166.7k-Neuron Fruit-Fly Connectome on One FPGA via Deferred Spike Expansion and Shared-Codeword States

YE Ziyang

cs.NEOct 28, 2026

FPGAFly: A 166.7k-Neuron Fruit-Fly Connectome on One FPGA via Deferred Spike Expansion and Shared-Codeword States

Authors: YE Ziyang

Organizations: Department of Electrical Engineering and Information Systems, Graduate School of Engineering, The University of Tokyo, Tokyo, Japan

Abstract

Synapse-resolution fruit-fly (Drosophila) connectomes enable neuron-by-neuron simulation of entire insect nervous systems, yet whole-brain models still require computationally demanding software or multiple neuromorphic chips, and single FPGAs support only small or statistical networks. Scaling to the full connectome, per-synapse spike expansion scales queue traffic with fan-out up to 11.2k, while 41-bit neuron states require 6.83 Mbit, exceeding the 6.27 Mbit on-chip block static RAM (BSRAM) of the target device. FPGAFly runs the 166.7k-neuron, 25.6M-connection leaky integrate-and-fire male central nervous system (CNS) connectome on one FPGA. Order-preserving deferred expansion queues one 64-bit descriptor per spike in source-ID order, cutting external memory traffic by 54.5%-57.9% and peak queue data by 95.6%-96.5% with bit-identical results. A 24-bit encoding shares codewords across normal, refractory, and quantized out-of-range states, reducing BSRAM usage from 438 to 270 with 1245 extra LUTs and zero changes across 10,000 MNIST predictions. At 80 MHz on a Gowin GW5AST-LV138, each 100 ms window finishes in 86.83 ms on average and within 93.05 ms across 2000 tests, realizing the first complete Drosophila CNS connectome simulation on one FPGA.

Figures & tables

Fig. 1: FPGAFly architecture. (a) System dataflow and execution phases. (b) Compute unit with parallel integration and serial event updates. (c) Shared 24-bit state encoding.

Variant	BSRAMs	LUTs	Registers	Setup WNS (ns)	Accuracy	Changed predictions	Rate MAE (Hz)	Population correlation	Spike match
	Synthesis			P&R at 80 MHz	Fidelity vs. uncompressed format
Uncompressed (41-bit)	438	37,953	56,328	exceeds capacity	86.91%	reference
Shared 24-bit (proposed)	270	39,198	48,272	+0.020	86.91%	0	0.0709–0.231	0.9904–0.9995	42.17%
Shared 24-bit, early expansion	270	39,859	48,127	+0.038	bit-identical to shared 24-bit on six inputs
Independent 29-bit	326	35,236	48,805	$-0.653$	86.92%	1	0.491–1.30	0.9868–0.9994	42.24%
Independent 24-bit	270	34,688	45,736	route failed	87.82%	987	9.64–18.5	$-0.0017$ –0.4053	1.06%

TABLE I: Resource Usage, Timing, and Fidelity of State-Format and Spike-Expansion Variants

Fig. 2: Early versus deferred spike expansion. (a) Record formats and dataflow. (b) Peak valid queue data and (c) DDR traffic per window.

Fig. 3: Board window timing and simulated per-step service time. (a) Window time over 2000 MNIST windows. (b) Maximum window time and overrun rate for synthetic inputs. (c) Window time with one white per nine black windows. (d) Per-step service time in RTL simulation over 3000 steps.

Statistical or procedural networks
Work	Platform	Devices	Network	Neurons	Connections	Time step	Real-time factor
Rhodes et al. [ 8 ]	SpiNNaker	Multi	Cortical microcircuit	77k	0.3B	0.1 ms	1.0
Kauth et al. [ 9 ]	Xilinx Virtex-7 XC7VX690T	35	Cortical microcircuit	77k	0.3B	—	0.05
R. Wang et al. [ 10 ]	Intel Stratix V 5SGXEA7	1	Cortex (pattern-based)	20M–2.6B	—	1 ms	1.0
Knight and Nowotny [ 11 ]	NVIDIA TITAN RTX	1	Macaque cortex (procedural)	4.13M	24.2B	—	—
Measured connectomes

TABLE II: Comparison With Prior Large-Scale Spiking-Network and Connectome Simulation Platforms

Spiking Neural Networks

Spiking neural networks process information through discrete, time-dependent neural events, aiming for efficient computation and biologically inspired learning.

See all 358 papers

A Connectome Test of the Fly Hashing Algorithm
Oct 6, 2026Sebastian SengeDimensionality Reduction Randomized Sketching
Independent Researcher
Sensorimotor behaviour from an untrained brain-and-nerve-cord connectome
Oct 24, 2026John R. Minnick, Jesus Gonzalez-Ferrer, Jinghui Geng +4Neural Processes Embodied Navigation
Department of Electrical and Computer Engineering, University of California, Santa Cruz, CA, USA. · UC Santa Cruz Genomics Institute, University of California, Santa Cruz, CA, USA. · Department of Biomolecular Engineering, University of California, Santa Cruz, CA, USA. +1
NeuroRing: Scaling Spiking Neural Networks via Multi-FPGA Bidirectional Ring Topologies and Stream-Dataflow Architectures
Apr 30, 2026Muhammad Ihsan Al Hafiz, Artur PodobasSpiking Neural Networks Neuromorphic Computing
KTH Royal Institute of Technology, Stockholm, Sweden

Oct 6, 2026J/K move · Enter open · S save

A Connectome Test of the Fly Hashing Algorithm

Sebastian Senge

Independent Researcher

Dimensionality Reduction Randomized Sketching Brain Network Analysis

Open paper AskPDF arXiv

Dasgupta, Stevens and Navlakha (2017) showed that the Drosophila olfactory circuit, modelled as a random sparse projection followed by winner-take-all, is a locality-sensitive hash that beats classical LSH. The projection was random because the wiring was unknown. We test it against four electron-microscopy connectomes (MaleCNS, hemibrain, FlyWire, BANC; four animals, seven hemispheres). First, the 2017 pattern holds in a reimplementation of its protocol on SIFT, MNIST and odour mixtures (GloVe is near chance at short codes for every method): the fly hash beats k Gaussian projections at short hash lengths (3.1x in AP@200 on MNIST at k = 4). Second, against the tested real-valued Gaussian baseline that advantage is per active cell, not per operation: Gaussian projections given the same projection arithmetic retrieve better on every dataset and input dimension tested. Third, across four connectomes the measured pairing of glomeruli gives no consistent retrieval advantage over degree-preserving rewiring: retrieval is slightly lower (median -1.6%), and the small odour deficits depend on how missing odour responses are treated. Separately, equalising glomerular fan-out at fixed connection count improves retrieval in the model in every hemisphere, while equalising inputs per cell lowers it on average. Yet the fan-out profile is similar across the four sampled animals (median between-animal Spearman rho = 0.86), structural synapse counts do not offset it, and its relation to odour tuning is weak. In the model that uneven allocation costs retrieval. For practice: measured wiring gives no consistent retrieval advantage over degree-preserving random wiring, so a fly hash needs no connectome data, and its advantage is per active unit, which may suit hardware where active units rather than arithmetic are the binding cost, a hypothesis we do not test.

Abstract

Spiking Neural Networks

Explore similar work

A Connectome Test of the Fly Hashing Algorithm

Sensorimotor behaviour from an untrained brain-and-nerve-cord connectome

NeuroRing: Scaling Spiking Neural Networks via Multi-FPGA Bidirectional Ring Topologies and Stream-Dataflow Architectures