Home

Organization

Mission Statement

Time Line

Research Interests

Neural Numbers

Scientific Advisory Committee

Useful Links

Research Topics of Interest

Multi-Petabyte Storage Systems

It has been variously estimated that the storage of uploaded information necessary for human brain emulation is in the neighborhood of 5-20 petabytes. Current reasonably priced (<$10,000) storage systems are in the vicinity of 1/1000th this capacity (Apple's X-RAID, 10.6 terabytes at $13K). The current trend is for this capacity to double every year and this appears to be an accelerating trend. Although it is not clear when this trend will run out of steam, current research indicates at least a ten fold increase is likely within the next three years. This would seem to verify that this trend will be upheld for at least the next ten years. That will put this technology in the ball park of what is necessary for neural uploading. Since this area is driven by computer market economics it is likely to take care of itself without added research funds or focus.

Understanding Neuro-Transmitters

There are currently dozens of neurotransmitters that have been identified that fall into three main categories: amines, amino acids and neuropeptides. In addition to these there are transmitters like nitric oxide (NO), and carbon monoxide(CO)that act in an area wide fashion. In addition, there are various growth factors that influence neural development. The full family of CNS neurotransmitters and other relevant brain chemicals still needs to be fleshed out and their activity fully characterized. This may require some funding outside of normal private industry, governmental and university channels.

Understanding Neural Types/Function

Basic CNS neural types include the fairly large and highly connected Purkinje calls and Pyramidal cells, and many other smaller cells with less connectivity like Basket cells, Stellate cells, Granule cells, and many other specialized cells. As in the neurotransmitter case the full characterization of these cells and other yet to be classified cells may require funding outside normal channels.

Brain Tissue Fixing and Microtome Development

Currently, people that are looking for an afterlife via technological advances are having their full bodies or heads frozen in liquid nitrogen by such organizations as Alcor or Cryonic Institute. Experimental evidence with large dog brains indicates considerable dessication and some cell damage due to ice crystalization. Better methods need to be developed for the preservation and preparation of brain tissue for subsequent scanning that may occur decades later. Research and funding for research in this area is not adequate but the problems are believed to be straight forward and should not require large scale funding resources.

Parallel Computer Systems Capable of Brain Emulation

Digital simulation of neurons require very fast floating-point calculations where weighting factors (degree of synaptic coupling) is multiplied by the input for each synapse and these added together for each neuron. That comes to 1,000 calculation per second for each synapse and an average of 1,000 synaptic junctions per neuron times 10¹¹ neurons yields a floor of 10¹⁷ floating-point multiply/adds (FMADs) per second. Computation resources capable of 10¹⁷ to 10¹⁸ FMADs per second are one to two orders of magnitude greater than super computers that exist today and maybe a perhaps a few orders of magnitude less expensive that IBM's "BlueGene" currently deployed using PowerPC microprocessors. Following Moore's law, the computational capability of microprocessors has been quadrupling every 18 months. The Law appears to be relatively sound for at least the next 10 years leading to the capability required. There are likely to be a few hiccups on the way such as the low K dielectric problem the industry is finally dealing with. Research in microprocessors and integrated circuit technology is largely done within companies or is company sponsored in the university environment.

Supercomputers like BlueGene (280TFlops/sec) are adequate for the job but not likely candidates for human brain emulation. First of all they will be far too expensive so it will take several more generations for that same capability to arrive at or near the desk top. Second, it may be of value to perform research into special purpose processors designed explicitly for neural processing since many simplifications are likely over using arrays of commercial general purpose microprocessors.

Highly Parallel Brain Tissue Recognition Equipment

This is likely the most difficult area of research. The resolution may require features smaller than 100 nanometers to be viewed which rules out ordinary optics for mapping the morphology of a neural network. Optics may be necessary for staining and recognition of larger features which can then be extended to help identify tinier features with devices such as TEMs (Transmission Electron Microscopes) or EUV microscopes. The suggested methodology would be that stacks of neural tissue would be aligned using blood capillaries as fiducial points. Neural networks would then be traced through these stacks. This methodology requires highly parallel automated scanning equipment and near real-time processing of neural tissue data. The storage of raw image data would require storage facilities several orders of magnitude greater than that anticipated for the storage of interconnection data. This is one area where we plan to concentrate research and research funding activity.

Copyright © 2003 LokTime Foundation. All Rights Reserved.