a
large-scale evolutionary hardware
research platform
Starlab, a multidisciplinary advanced research institute now based in Barcelona, Spain, focuses on advanced scientific research projects with significant long-term potential impact for humanity.
My research group was tasked with designing and evolving artifical neural networks directly in silico, combining evolvable hardware and genetic algorithms in a novel approach towards the design and implementation of large-scale neural networks.
Our custom hardware could simultaneously update 131 Billion cells per second.
Computational
Power
~ 10,000 Pentium III PCs
Cellular Automata Update Rate
131 billion cells / second
Number
of Supported Neurons
74.5 million
Number
of Supported Cells
893 million
Number
of neurons per neural module
1152
Number
of Supported Neural Modules
64,640
Neural
Module Chromosome Length
91,000 bits
Number
of FPGAs
72 Xilinx XC-6264BG560
Number
of FPGA Reconfigurable Function Units
1,179,648
Phenotype
/ Genotype Memory
1.18 Gbytes
Power Consumption
1.5 KWatt [ 5V, 300A ]