Neuromorphic Electronics - Forsiden

1 NANON anoelectronicsNeuromorphic ElectronicsLecture NotesFall Term 2010 Department of InformaticsUniversity of Osloread byPhilipp D. H afligerAbstractThis is the script for an introductory course in Neuromorphic electronic circuits. These are circuitsinspired by the nervous system that either help verifying neuro-physiological models, or that areuseful components in artificial perception/action systems. Research also aims at using them inimplants. These circuits are computational devices and intelligent sensors that are very differentlyorganized than digital processors. Their storage and processing capacity is distributed. They areasynchronous and use no clock signal. They are often purely analog and operate time are adaptive or can even learn on a basic level instead ofbeing programmed. A shortintroduction into the area of brain research is also included in the students will learn to exploit mechanisms employed by the nervous system for compactenergy efficient analog integrated circuits.

2 They will get insight into a multidisciplinary researcharea. The students will learn to analyze analog CMOS circuits and acquire basic knowledge inbrain research script is not really a book yet and many of the explanations are brief and not very script is intended as supportive material for the courseand may be quite unsatisfying to readwithout attending the lectures. The students are strongly encouraged to add their own notes tothese Neuromorphic Circuits at Present .. 12 Neurophysiology in a Methods .. Psychophysical Experiments .. EEG .. fMRI and PET .. Extracellular Electrodes .. Intracellular Electrodes .. Fluorescent Tracers and Imaging .. Briefly Mentioned: Methods in Neuroanatomy .. Knowledge .. Brain Anatomy .. Cortical Regions .. Organization within Cortical Regions.

3 Microcolumns and Cortical Layers .. Neurons and Synapses .. 183 Basic Analog Field Effect Transistors .. Basic Formulae .. Early effect .. Gate leakage .. Capacitors .. Current Mirror .. Differential Pair .. Transconductance Amplifier .. Follower .. Resistor .. Resistive Nets .. The Winner Take All circuit .. 364 Real and Silicon Real Neurons .. aVLSI Models of Neurons .. Simple Electrical Nodes as Neurons .. Perceptrons (Mc Culloch Pitts neurons) .. Integrate and Fire Neurons .. Compartemental Neuronal Models (Silicon Neurons) .. 525 Coding in the Nervous The action potential .. Hints in Experiments .. Classical experiments based on observing spike rates.. Classical Experiments observing temporal spike codes .. Candidate Codes.

4 636 Neuromorphic Communication: the AER The Basic Idea of Address Event Representation (AER) .. Collision Handling .. Full Arbitration .. Collision Discarding .. Aging versus Loss Trade Off .. 747 Retinomorphic The Retina .. CMOS photo sensors .. Photo diodes .. Photo transistors .. Photo gates .. Photo Current Amplification .. Linear by Early effect .. Logarithmic by gate to source voltage .. Common source amplification .. Source follower .. Read Out Strategies .. Addressing and scanning .. Charge coupled devices (CCD) .. Address event representation .. Silicon retinae .. Adaptive photo cell .. Spatial contrast retina .. Temporal contrast retina .. Further Image Processing .. Motion .. Feature maps.

5 1008 Cochleomorphic The Cochlea .. Silicon Cochlea .. 1119 Neuromorphic Neural Learning Algorithms .. An overview of classes of learning algorithms .. Supervised Learning .. Reinforcement learning .. Unsupervised learning .. Analogue Storage .. Dynamic Analogue Storage .. Static Analogue Storage .. Non-Volatile Analogue Storage .. Neuromorphic Learning Circuits .. Hebbian learning circuits .. A spike based learning circuit .. 141A Questions Introduction .. Neurophysiology .. Basic Analogue CMOS .. Real and Silicon Neurons .. Coding in the Nervous System .. Neuromorphic Communication: the AER Protocol .. Retinomorphic Circuits .. Cochleomorphic Circuits .. Neuromorphic Learning .. 147viCONTENTSList of EEG array based reconstruction of brain activity.

6 EEG of sleep stages .. fMRI in bilingual task .. The Utah electrode array .. Patch clamp electrodes .. Fluorescent tracers in Purkinje cell .. Brain Cross section illustration .. Motor-cortex homunculus .. Brain research by Garry Larson .. Cortical regions in a cat .. Cortical regions hierarchy .. ocular dominance patterns on V1 .. ocular dominance patterns close-up .. Orientation selectivity patterns .. Cortical layers staining illustration 1 .. Cortical layers staining illustration 2 .. Cortical layers staining techniques .. Schematic synapse .. FETs .. Capacitances in CMOS .. Current mirror .. Differential pair .. Transconductance amplifier .. Follower .. Resistors in CMOS .. Resistive net .. Diffuser net.

7 WTA principle .. CMOS WTA .. WTA with spatial smoothing .. local WTA .. WTA with hysteresis .. Anatomical Parts of a Neuron .. Light Microscope Neuron .. 3D Reconstruction of Pyramidal Cell .. 45viiviiiLIST OF Perceptron Concept .. Perceptron Schematics .. Gilbert Multiplier .. Concept Integrate-and-Fire Neuron .. Carver Mead Integrate-and-Fire Neuron .. Adaptive Integrate-and-Fire Neuron .. Compartemental Neuron Model .. The Hudgkin Huxley Model .. A CMOS Implementation of a HH-soma .. Cable Model .. Galvani experiment with twitching frog legs .. Orientation selective neuron responses .. Exact responses to random dot patterns 1 .. Synfire chains .. Phase relation of place cells .. Illustration of coding schemes .. Latency coding.

8 Address Event Representation .. 4 Phase Handshake .. two-input greedy arbiter .. Glitch free two-input greedy arbiter .. Binary arbiter tree .. AER with collision discarding .. Aging versus Loss trade off AER .. Eyball cross section .. Detailed retinal cells .. Schematic retinal cells .. Photo diode .. Photo diode layout .. PNP photo transistor .. Photo gate .. Amplification by drain resistance .. Logarithmic amplification .. Two transistor inverting amplifier .. Negative feedback .. Active pixel .. CCD .. AER photo pixel .. Adaptive photo cell .. Non linear element .. Mahowald silicon retina .. Boahen silicon retina .. Temporal contrast retina diagram .. Temporal contrast retina transistor level circuit .. Reichardt detector.

9 Intensity based motion estimation .. Original natural scene .. Surface plot of 2D difference of Gaussians .. Colour code plot of 2D difference of Gaussians .. 104 LIST OF Difference of Gaussians convolved image .. Surface plot of a 45 degree edge extraction kernel .. Colour code plot of a 45 degree edge extraction kernel .. Image after 45% edge extraction .. Ear cross section .. Cochlea cross section .. EM of hair cells .. A second order filter stage .. Parallel second order filter spectra .. Cascaded second order filter spectra .. Character recognition by associative memory .. Classification with LVQ .. Dynamics in Learning Vector Quantization .. Dynamics in competitive Hebbian learning .. competitive learning vs. associative memory .. Spike based learning in a silicon neuron.

10 Spike based Learning simulation variables .. Capacitive dynamic analog storage .. AD/DA multi-level static storage .. A fusing amplifier .. Weak multi-level static memory .. Floating gate, non-volatile analog storage .. Band diagram for tunneling through the gate oxide .. High voltage NFET .. On chip high voltage switch .. Diorio learning array .. Fusi bistable learning circuit .. Blockdiagram of a spike based learning circuit .. Positive term circuit .. Negative term circuit .. Floating gate analog storage cell .. 143xLIST OF FIGURESC hapter 1 IntroductionThe term Neuromorphic was introduced by Carver Mead around 1990. He defined neuromorphicsystems as artificial systems that share organization principles with biological nervous system. Sowhat are those organization principles?A brain is fundamentally differently organized than a computer and science is still a long wayfrom understanding how the whole thing works.