micromachines Article
Microfluidic Neurons, a New Way in Neuromorphic Engineering? Timothée Levi * and Teruo Fujii LIMMS (Laboratory for Integrated Micro Mechatronic Systems)/CNRS-IIS (Institute of Industrial Science), The University of Tokyo, 4-6-1 Komaba, Meguroku, 153-8505 Tokyo, Japan;
[email protected] * Correspondence:
[email protected]; Tel.: +33-5-4000-3118 Academic Editor: Nikos Chronis Received: 20 July 2016; Accepted: 18 August 2016; Published: 22 August 2016
Abstract: This article describes a new way to explore neuromorphic engineering, the biomimetic artificial neuron using microfluidic techniques. This new device could replace silicon neurons and solve the issues of biocompatibility and power consumption. The biological neuron transmits electrical signals based on ion flow through their plasma membrane. Action potentials are propagated along axons and represent the fundamental electrical signals by which information are transmitted from one place to another in the nervous system. Based on this physiological behavior, we propose a microfluidic structure composed of chambers representing the intra and extracellular environments, connected by channels actuated by Quake valves. These channels are equipped with selective ion permeable membranes to mimic the exchange of chemical species found in the biological neuron. A thick polydimethylsiloxane (PDMS) membrane is used to create the Quake valve membrane. Integrated electrodes are used to measure the potential difference between the intracellular and extracellular environments: the membrane potential. Keywords: biomimetic artificial neuron; microfluidic; action potential
1. Introduction Millions of people worldwide are affected by neurological disorders which disrupt connections between brain and body, causing paralysis or affecting cognitive capabilities. The number is likely to increase over the next few years and current assistive technology is still limited. In recent decades, extensive research has been devoted to brain-machine interfaces (BMIs), and neuroprostheses in general [1–3], working towards effective treatment for these disabilities. The development of these devices has had and, hopefully, will continue to have a profound social impact on these patients' quality of life. These prostheses are designed on the basis of our knowledge of interactions with neuronal cell assemblies, taking into account the intrinsic spontaneous activity of neuronal networks and understanding how to stimulate them into a desired state or produce a specific behavior. The long-term goal of replacing damaged neural networks with artificial devices also requires the development of neural network models matching the recorded electrophysiological patterns and capable of producing the correct stimulation patterns to restore the desired function. The hardware setup used to interface the biological component is a biomimetic neural network system implementing biologically realistic neural network models, ranging from the electrophysiological properties of a single neuron to large-scale neural networks. This research field is named “neuromorphic engineering”. The neuromorphic engineering is a new emerging interdisciplinary field merging biology, physics, mathematics, computer sciences, and engineering approaches to design biomimetic artificial neural systems. These artificial neural networks emulate the electrical activity of biological neural networks and their goal is to mimic and/or replace the organic ones. Most of these systems are silicon-based [4,5]. The main goals of those systems are the design of tools for biomedical applications, Micromachines 2016, 7, 146; doi:10.3390/mi7080146
www.mdpi.com/journal/micromachines
2016, 7, 146 Micromachines 2016,
2 of 12
based [4,5]. The main goals of those systems are the design of tools for biomedical applications, like like neuroprostheses understanding humannervous nervoussystem system[7,8]. [7,8].The Theuse use of of silicon silicon neuroprostheses [6], [6], andand thethe understanding of of thethe human neurons brings some bio-compatibility issues (rejection, power consumption). To circumvent neurons brings some bio-compatibility issues (rejection, power consumption). To circumvent those those issues, issues, aa new new way way of of neuromorphic neuromorphic engineering engineering should should be be explored: explored: the the artificial artificial neural neural systems systems based on microfluidic microfluidictechniques techniques(Figure (Figure1).1).The The aim is design to design a neuromimetic architecture of based on aim is to a neuromimetic architecture of one one neuron based on the use of microfluidic techniques [9] and microsensor integration. In Figure 1, neuron based on the use of microfluidic techniques [9] and microsensor integration. In Figure 1, the the microfluidic neuron is between the silicon neuron andbiological the biological It could good microfluidic neuron is between the silicon neuron and the one. Itone. could offer aoffer goodatradetrade-off with more biological plausibility than the silicon neuron, and to make hybrid experiments in off with more biological plausibility than the silicon neuron, and to make hybrid experiments in an an easier manner (bi-directional communication betweena aliving livingneuron neuronand andan anartificial artificial one). one). To To our our easier manner (bi-directional communication between knowledge, knowledge, this this new new methodology methodology does does not not exist exist yet yet in in the the state state of of the the art. art.
Figure 1.1.Microfluidic Microfluidic neuron position the biocompatibility neuron modelling for hybrid Figure neuron position in theinbiocompatibility neuron modelling for hybrid experiments. experiments.
In this article, article, we we will will first first describe describe the novelty and and the the state state of of the the art art about about biomimetic biomimetic artificial artificial In this the novelty neurons, microfluidic neuron neuron modelling modelling and and its its design; design; finally, finally, we neurons, then then discuss discuss microfluidic we will will present present the the first first results and perspectives of this work. results and perspectives of this work. 2. Novelty and Comparison with the State of the Art in Silicon Neuron 2. Novelty and Comparison with the State of the Art in Silicon Neuron 2.1. Silicon Neuron for Hybrid Neural Network 2.1. Silicon Neuron for Hybrid Neural Network In neuromorphic engineering, the most commonly used systems are silicon neurons [10]. Silicon In neuromorphic engineering, the most commonly used systems are silicon neurons [10]. Silicon neurons are hybrid analog/digital very large scale integration (VLSI) circuits that emulate the neurons are hybrid analog/digital very large scale integration (VLSI) circuits that emulate the electrophysiological behavior of real neurons and their conductances. Silicon neuron circuits represent electrophysiological behavior neurons and their conductances. Silicon neuron circuits one of the main building blocks of forreal implementing neuromorphic systems. The term “neuromorphic” represent one of the main building blocks for implementing neuromorphic systems. The term was coined by Carver Mead [11] in the late 1980s. It refers to artificial neural systems whose architecture “neuromorphic” was coined by Carver Mead [11] in the late 1980s. It refers to artificial neural systems and design principles are based on those of biological nervous systems. The term “neuromorphic” whose architecture design VLSI principles arethat based on those ofthe biological nervous The used term also describes the setand of analog circuits operate using same physics of systems. computation “neuromorphic” also describes the set of analog VLSI circuits that operate using the same physics of by the nervous system. It implies that the artificial neuron circuit exploits the physics of the silicon computation used by the nervous system. It implies that the artificial neuron circuit exploits the medium to directly reproduce the bio-physics of nervous cells. physics of the silicon directlythe reproduce of anervous cells. In designing an medium artificial to neuron, first stepthe is bio-physics the choice of biologically-realistic model In designing an artificial neuron, the first step is the choice of a biologically-realistic model (Figure 2). Indeed, a mathematical model based differential equations is capable of reproducing a (Figure 2). Indeed, a mathematical model based differential equations is capable of reproducing a behavior quite similar to that of a biological cell. The choice of model is based on two criteria: the family behavior quite similar to that of a biological cell. The choice of model is based on two criteria: the of neurons able to be reproduced and the number of equations. Models could mimic from complex family of neurons be reproduced and the number of equations. [12]) Models mimic from biophysical modelsable that to emulate ion channel dynamics (Hodgkin-Huxley andcould detailed dendritic complex models thatintegrate-and-fire emulate ion channel or axonalbiophysical morphologies to basic [13]dynamics circuits. (Hodgkin-Huxley [12]) and detailed dendritic or axonal morphologies to basic integrate-and-fire [13] neuron circuits.circuits can be more or less Depending on the application domain of interest, artificial complex, with large arrays of neurons all integrated on the same chip [14–16], or single neurons implemented on a single chip [17], or with some elements of the neuron distributed across multiple chips [18].
Micromachines 2016, 7, 146
3 of 12
Micromachines 2016, 7, 146 Micromachines 2016, 7, 146
3 of 12 3 of 12
Figure 2. Neuron models in the literature classified following the biological plausibility and the implementation cost. The figure was extracted from www.izhikevich.com. An electronic version of the figure and reproduction permissions are freely available at www.izhikevich.com Figure 2. 2. Neuron Neuron models in the the literature literatureof classified following the biological plausibility and the the Figure in classified biological plausibility and Depending on themodels application domain interest,following artificialthe neuron circuits can be more or less implementation cost. The figure was extracted from www.izhikevich.com. An electronic version of implementation cost. The figure was extracted from www.izhikevich.com. An[14–16], electronicor version the complex, with large arrays of neurons all integrated on the same chip singleofneurons the figure and permissions are freely available at the www.izhikevich.com figure andon reproduction permissions freely available at www.izhikevich.com. implemented areproduction single chip [17], or are with some elements of neuron distributed across multiple
chips [18]. Depending on the application domain of interest, artificial neuron circuits can be more or less However, of them [19–23] allow allow hybrid hybrid experiments with living neuron cells.neuron The specifications However,few few of them [19–23] experiments with living cells. The complex, with large arrays of neurons all integrated on the same chip [14–16], or single neurons of such systems, real-time with biological time-scale, using complex neuron models specifications of are: suchworking systems,inare: working in areal-time with a biological time-scale, using complex implemented on a single chip [17], or with some elements of the neuron distributed across multiple that mimic the spike timing,the and/or morphology of morphology action potential. neuron models that mimic spike the timing, and/or the of action potential. chips [18]. Figure 33 describes describesone onehybrid hybrid experiments the silicon neuron can interact with experiments andand howhow the silicon neuron can interact with living However, few of them [19–23] allow hybrid experiments with living neuron cells. The living neuronal A neuron model (Hodgkin-Huxley formalism) implementedinto into the the ASIC neuronal cells. cells. A neuron model (Hodgkin-Huxley formalism) is isimplemented specifications of such systems, are: working in real-time with a biological time-scale, using complex (Application-Specific Integrated Circuit) Circuit) which which controlled controlled the the synaptic synaptic current current applied applied to to “in “in vitro” vitro” neuron models that mimic the spike timing, and/or the morphology of action potential. neuron experiments where thethe neural activity is recorded and neuron culture. culture.There Thereare arealso alsosome someclosed-loop closed-loop experiments where neural activity is recorded Figure 3 describes one hybrid experiments and how the silicon neuron can interact with living controlled the artificial neural network [24]. [24]. and controlled the artificial neural network neuronal cells. A neuron model (Hodgkin-Huxley formalism) is implemented into the ASIC (Application-Specific Integrated Circuit) which controlled the synaptic current applied to “in vitro” neuron culture. There are also some closed-loop experiments where the neural activity is recorded and controlled the artificial neural network [24].
Figure 3. 3. AAhybrid neuron using Hodgkin-Huxley formalism andand “in Figure hybridexperiment experimentbetween betweena silicon a silicon neuron using Hodgkin-Huxley formalism vitro” neurons with intracellular recording [5]. V(I syn) is the voltage associated to synaptic current, “in vitro” neurons with intracellular recording [5]. V(Isyn ) is the voltage associated to synaptic current, Vmem_post isisthe V themembrane membranevoltage voltageofofthe thepost-synaptic post-synapticneuron. neuron. mem_post Figure 3. A hybrid experiment betweenthe a silicon neuron using Hodgkin-Huxley formalism and “in These silicon neurons reproduce time-spiking of the neural network but their power These silicon neurons reproduce the time-spiking of the neural network but their power vitro” neurons with intracellular recording [5]. V(Isyn) is of theoperating voltage associated to synaptic current, consumption is high. Another drawback is the difficulty hybrid experiments as it needs consumption Another drawback is the difficulty of operating hybrid experiments as it isis thehigh. membrane of the post-synaptic neuron.detection, Vmem_post different electrical modulesvoltage like stimulation, recording, amplification, and filtering needs different electrical modules like stimulation, recording, detection, amplification, and filtering modules. Several neurotechnologies for interfacing with “in vitro” neurons are already described in modules. for interfacing with “in neurons are already in TheseSeveral siliconneurotechnologies neurons reproduce the time-spiking ofvitro” the neural network but described their power the literature but these system are for stimulation and/or recording of spiking activities [25–28], or the literature but theseAnother system are for stimulation and/or recording of spiking activities [25–28], or for consumption is high. drawback is the difficulty of operating hybrid experiments as it needs for drug screening and neuroscience research [29–31]. They are not designed to mimic the living part drug screening and neuroscience research [29–31].recording, They are not designedamplification, to mimic the living and different electrical modules like stimulation, detection, and part filtering and to replace it in the case of hybrid experiments. to replace Several it in the neurotechnologies case of hybrid experiments. modules. for interfacing with “in vitro” neurons are already described in
the Microfluidic literature but these for system areNeural for stimulation 2.2. Neuron Hybrid Network and/or recording of spiking activities [25–28], or 2.2. Microfluidic Neuron for Hybrid Neural Network for drug screening and neuroscience research [29–31]. They are not designed to mimic the living part and to replace in the caseofofthe hybrid experiments. The main it advantage microfluidic neuron is the biocompatibility. We can design in one single chip, the microfluidic neuron and the living neuron (Figure 4). Polydimethylsiloxane (PDMS) is 2.2. Microfluidic Neuron for Hybrid Neural Network
Micromachines 2016, 2016, 7, 7, 146 146 Micromachines
of 12 12 44 of
Micromachines 2016, 7, 146
4 of 12
The main advantage of the microfluidic neuron is the biocompatibility. We can design in one single chip, the microfluidic neuron and the living neuron (Figure 4). Polydimethylsiloxane (PDMS) made of biocompatible material andand aqueous KCl KCl and NaCl solutions are used creating the action is made of biocompatible material aqueous and NaCl solutions are for used for creating the potential like in living action potential like inneurons. living neurons.
Figure 4. 4. A A single single chip chip representation representation of of aa hybrid hybrid experiment experiment using using microfluidic microfluidic neuron. neuron. Neuron Neuron Figure stimulation is made from a microfluidic neuron electrode. The control of the microfluidic neuron stimulation is made from a microfluidic neuron electrode. The control of the microfluidic neuron valve valve is done by the electrical activity of “in vitro” neurons. V mem is the action potential of the is done by the electrical activity of “in vitro” neurons. Vmem is themem action potential of the biological microflu is the action potential of the microfluidic neuron. biological microflu neuron, V neuron,isVthe action potential of the microfluidic neuron. microflu
The interactions between them use the voltage membrane for stimulating each neuron. As the The interactions between them use the voltage membrane for stimulating each neuron. As the microfluidic neuron is using the same ionic channel than the living one, the feasibility of hybrid microfluidic neuron is using the same ionic channel than the living one, the feasibility of hybrid experiments increases. experiments increases. 3. Design 3. Designof ofthe theMicrofluidic Microfluidic Neuron Neuron 3.1. Neuron Neuron Modelling Modelling 3.1. The biological biological neuron neuron transmits transmits electrical electrical signals signals based based on on ion ion flow flow through through their their plasma plasma The membrane. The interior of the the neurons neurons has has aa negative The action action negative potential, potential, called called resting resting potential. potential. The membrane. potential temporarily abolishes the resting potential negative and positive, creating the potential temporarily abolishes the resting potential negative and positive, creating the transmembrane transmembrane potential. Action potentials are propagated along axons and represent the potential. Action potentials are propagated along axons and represent the fundamental electrical fundamental electrical signals by information transmitted from oneinplace anothersystem. in the signals by which information is which transmitted from is one place to another the to nervous nervous system. Ion channels have an activity dependent on the potential difference between the Ion channels have an activity dependent on the potential difference between the intracellular and intracellular and extracellular environments. The properties of these channels give rise to an electrical extracellular environments. The properties of these channels give rise to an electrical phenomenon phenomenonalong propagating the axon: the action potential (Figure 5). propagating the axon:along the action potential (Figure 5).
Figure 5. 5. Description of of action potential following the the different ion ion interactions. (A) Figure Descriptionofofthe thecreation creation action potential following different interactions. + voltage steady state of membrane voltage; (B)(B) the (A) steady state of membrane voltage; theopening openingofofNa Na+ +ionic ionicpump pump depolarizes depolarizes the voltage membrane; (C) closing closing of of Na Na+++ ionic pump; (C’) the opening of K K+++ ionic pump repolarizes the voltage membrane; (D) (D) come come back back to to the the resting resting potential potential thanks thanks to to leakage leakage current. current. membrane;
computation cost. The most neuromimetic model, and also the most complex, is the Hodgkin-Huxley formalism [12]. In this model, an electrical circuit has the same electrical behavior of the membrane. The different branches that make up this circuit are the membrane capacitance between extra-and intracellular environments, current generators for sodium (Na) and potassium (K) that have voltagedependent conductances and, finally, a constant leakage channel conductance. This formalism is Micromachines 2016, 7, 146 5 of 12 widely used in neuroscience as it is generalizable to any type of nerve cell. In silicon neurons, this model is hard to implement as the computation cost is prohibitive. However using microfluidic As seenthe previously, there are is different thethis bioplausibility the techniques, implementation easier neuron than formodels silicondepending ones. We of use formalism and for our computation cost. The most neuromimetic model, and also the most complex, is the Hodgkin-Huxley microfluidic device. formalism [12]. In this model, an electrical circuit has the same electrical behavior of the membrane. 3.2. Microfluidic Neuron that make up this circuit are the membrane capacitance between extra-and The different branches intracellular environments, current generators for sodium (Na) and potassium (K) that have The main goals of this artificial neuron are to simplify the interactions with biological neurons, voltage-dependent conductances and, finally, a constant leakage channel conductance. This formalism to reduce the bio-compatibility issues keeping the same ionic currents (K+, etc.) of biological neurons is widely used in neuroscience as it is generalizable to any type of nerve cell. In silicon neurons, this and to use biocompatible materials, like PDMS. model is hard to implement as the computation cost is prohibitive. However using microfluidic Based on the physiological behavior of the biological neuron (Figure 5) and Hodgkin-Huxley techniques, the implementation is easier than for silicon ones. We use this formalism for our formalism [12], we propose a microfluidic structure composed of chambers representing the intra microfluidic device. and extracellular environments, connected by channels actuated by Quake [32] valves. These channels are equipped 3.2. Microfluidic Neuron with selective permeable membranes (Nafion) to mimic the exchange of species found in the biological neuron. In our case (Figure 6), the K+ is the only cation that can go The the main goals of this artificial neuron are to simplify biological neurons, to through membrane, which generates a potential when the the interactions electrons gowith the opposite way through + , etc.) of biological neurons and reduce the bio-compatibility issues keeping the same ionic currents (K the gold electrodes. The intra and extracellular chambers are controlled by pressure to control the to use biocompatible materials, PDMS. exchange due to convection. A like network of integrated gold microelectrodes is used to measure the Based on the physiological behavior of the biological neuron (Figure 5) and Hodgkin-Huxley potential difference between the intracellular and extracellular environments: the membrane formalism [12], we propose a microfluidic structure composed of chambers representing the intra and potential. extracellular environments, connectedionic by channels actuated by Quake(Nernst [32] valves. These thanks channels A fixed potential using different concentrations is generated equations) to are equipped with selective permeable membranes (Nafion) to mimic the exchange of species found a selective membrane. A controlled valve is also used to modify this fixed voltage potential which in the biological neuron. Inartificial our caseaction (Figure 6), the K+ is the only cation that can go through the allows the generation of the potential. membrane, which generates a potential when the electrons goKCl the opposite way through the gold In our experiment (Figure 6), two chambers with different concentrations are separated by electrodes. The intra and extracellular chambers are controlled by pressure to control the exchange a selective membrane: the Nafion membrane [33]. C+ corresponds to high concentration, C− to low due to convection. A network of integrated goldasmicroelectrodes is used measure the potential concentration. The voltage potential is generated only positive ions (K+ intoour experiment) can go difference between the intracellular and extracellular environments: the membrane potential. through the Nafion membrane.
Figure in the the microfluidic microfluidicneuron. neuron.CC++ corresponds corresponds to to high high concentration, concentration,CC−− to Figure 6. 6. Ionic Ionic exchanges exchanges in to low low concentration. The electrical potential form the anode and cathode mimics the action potential of concentration. The electrical potential form the anode and cathode mimics the action potential of one one − neuron. membrane allows allows the the flow flow of of K K++ but neuron. The The Nafion Nafion membrane but reject reject the the negative negative ion, ion, like like OH OH−. .
We generated a voltage potential between the two chambers. To modify this potential, a A fixed potential using different ionic concentrations is generated (Nernst equations) thanks to controlled valve is used to close the channel (Figure 7). If the valve restricts 100% of the interchamber a selective membrane. A controlled valve is also used to modify this fixed voltage potential which flow, there is no more conduction and the voltage potential is null. However, as our air-pressure allows the generation of the artificial action potential. controlled valve cannot be fully closed, the voltage potential can be controlled by the air pressure In our experiment (Figure 6), two chambers with different KCl concentrations are separated by value. a selective membrane: the Nafion membrane [33]. C+ corresponds to high concentration, C− to low concentration. The voltage potential is generated as only positive ions (K+ in our experiment) can go through the Nafion membrane. We generated a voltage potential between the two chambers. To modify this potential, a controlled valve is used to close the channel (Figure 7). If the valve restricts 100% of the interchamber flow, there is no more conduction and the voltage potential is null. However, as our air-pressure controlled valve cannot be fully closed, the voltage potential can be controlled by the air pressure value.
Micromachines 2016, 7, 146 Micromachines 2016, 7, 146 Micromachines 2016, 7, 146
6 of 12 6 of 12 6 of 12
Figure 7. Role of the valve for the voltage potential. The open valve keeps constant voltage potential from the two chambers. The closed valve controls the amplitude of the voltage potential depending on the air-pressure used.
A concentration polarization is created thanks to the valves due to nanochannel modelling [34]. This phenomena seems to appear in our case and explain the variation of potential depending of the Figure 7. Roleofofthe thechannel. valve for the voltage potential. The open valve keeps constant voltage potential variable leakage from the The valve potential. controls the amplitude of keeps the voltage potential depending Figure 7. two Role of valve forclosed the voltage voltage potential. The open constant voltage potential Figure 7. Rolechambers. of the the valve for the The open valve valve keeps constant voltage potential on the air-pressure used. from the two chambers. The closed valve controls the amplitude of the voltage potential depending 3.3. Design Steps from the two chambers. The closed valve controls the amplitude of the voltage potential depending on on air-pressure used. thethe air-pressure used.
To design this system, severalissteps are thanks needed.to the valves due to nanochannel modelling [34]. A concentration polarization created In Figure 8, the left chamber the (high concentration of[34]. K+) This A phenomena seems to appearrepresents in our casethanks andintracellular explain the environment variation potential depending the concentration polarization is created to the valves due toof nanochannel modellingof polarization is created(low thanks to the valves nanochannel modelling [34]. and A theconcentration right chamber, extracellular ofdue K+).toElectrodes inside the variable leakage of the channel. This phenomena seems tothe appear in our case andconcentration explain the variation of potentialplaced depending of the This phenomena seems to appear in our case and explain the variation of potential depending of the chambersleakage monitor variable of the the potential channel. difference between the two fields: the membrane potential. variable leakage the channel. Exchanges between the extracellular and intracellular chambers are done through 3.3. Design Steps of microchannels equipped with a selective membrane (Nafion) and controllable by Quake valves, as 3.3. Steps 3.3. Design Design Stepsthis system, several steps are needed. To design described in Figure 9. These valves are electronically controlled to reproduce the Hodgkin-Huxley In Figure 8,this thesystem, left chamber represents the intracellular environment (high concentration of K+) To design several steps are needed. needed. To design thisthe system, several are formalism. When Quake valvessteps are activated, the channel is closed and the membrane potential and In the right 8, chamber, the extracellular concentration of K+). Electrodes placed inside the the left left chamber chamber represents(low the intracellular intracellular environment (high concentration concentration of K K ++)) In Figure Figure represents the environment (high of is equal to 0 V.8, the + chambers monitor the potential difference between the two fields: the membrane potential. and the right chamber, thedesigned extracellular (low concentration of ).µm. Electrodes placed the + ). and the chamber, the extracellular concentration Electrodes placed inside inside the The right Quake valves are using(low a PDMS membraneof ofKK 15 The air pressure channels Exchanges between the difference extracellular andtheintracellular chambers are done through chambers monitor the potential between two fields: the membrane potential. chambers difference two fields: the membrane potential. are 20 µm monitor × 50 µm.the Thepotential liquid channels arebetween 10 µm ×the 10 µm. microchannels membrane and controllable by Quake as Exchangesequipped betweenwith thea selective extracellular and (Nafion) intracellular chambers are done valves, through described in Figure 9. These are electronically controlled tocontrollable reproduce the microchannels equipped withvalves a selective membrane (Nafion) and by Hodgkin-Huxley Quake valves, as formalism.inWhen the9.Quake the channel is closed and thethe membrane potential described Figure These valves are are activated, electronically controlled to reproduce Hodgkin-Huxley is equal to 0When V. the Quake valves are activated, the channel is closed and the membrane potential formalism. ThetoQuake is equal 0 V. valves are designed using a PDMS membrane of 15 µm. The air pressure channels are 20 µmQuake × 50 µm. The are liquid channels are a10PDMS µm × 10 µm. The valves designed using membrane of 15 µm. The air pressure channels are 20 µm × 50 µm. The liquid channels are 10 µm × 10 µm. Figure 8. Microfluidic neuron device. device. Figure 8. Microfluidic neuron
Exchanges between the extracellular and intracellular chambers are done through microchannels equipped with a selective membrane (Nafion) and controllable by Quake valves, as described in Figure 9. These valves are electronically controlled to reproduce the Hodgkin-Huxley formalism. When the Quake valves are activated, the8.channel is closed and the membrane potential is equal to 0 V. Figure Microfluidic neuron device. Figure 8. Microfluidic neuron device.
Figure 9. Quake valve off and on. Fluidic channel is 10 µm × 10 µm in size. The Quake valve is 20 µm × 50 µm in size. The end of Quake valve is a 1 mm diameter circle.
The Quake valves are closed using air pressure. Depending of the value of the pressure, the closing time constant can be adjusted. When the air-pressure is stopped, the valve is opened with a time constant depending of the diffusion time of the ionic solutions for recovering the ionic flow. As Figure 9.the Quake off and andfor on.KFluidic Fluidic channel is 10 10 µm µm × 10 µm Quake is + and Cl − are nearly Figure Quake valve off on. channel is × 10 µm in in size. The Quake valve is 20 20 µm µm KCl is used,9. timevalve diffusion the same: K+size. 64.7The cm/s, Cl− valve 65.4 cm/s. ×The 50 in size. The end of Quake valve is a 1 mm diameter circle. Figure 9. Quake valve off and on. Fluidic channel is 10 µm × 10 µm in size. The Quake valve is 20 µm × 50µm µm in size. The end of Quake valve is a 1 mm diameter circle. selective Nafion membrane is 8 µm wide, located between the two PDMS layers of the × 50 µm inNafion size. Themembrane end of Quake valve is a 1with mm Nafion diameterperfluorinated circle. channels. The is designed resin solution 5 wt % using The Quake valves are closed using air pressure. Depending of the value of the pressure, the The Quake valves are designed using a PDMS membrane of 15 µm. The air pressure channels are closing constant can adjusted. When the air-pressure is stopped, valve is opened with Thetime Quake valves arebeclosed using air pressure. Depending of the the value of the pressure, thea 20 µm × 50 µm. The liquid channels are 10 µm × 10 µm. time constant depending diffusion timethe of the ionic solutions for recovering ionic flow. closing time constant can of bethe adjusted. When air-pressure is stopped, the valvethe is opened withAs a KCl isconstant used, the time diffusion K+ and time Cl− are thesolutions same: K+for 64.7 cm/s, Cl− the 65.4ionic cm/s.flow. As time depending of thefor diffusion of nearly the ionic recovering − are − 65.4 cm/s. The selective Nafion membrane is 8Clµm wide, located between thecm/s, twoCl PDMS layers of the KCl is used, the time diffusion for K+ and nearly the same: K+ 64.7 channels. The Nafion membrane is designed with Nafion perfluorinated resin wt % of using The selective Nafion membrane is 8 µm wide, located between the twosolution PDMS 5layers the channels. The Nafion membrane is designed with Nafion perfluorinated resin solution 5 wt % using
Micromachines 2016, 7, 146
7 of 12
The Quake valves are closed using air pressure. Depending of the value of the pressure, the Micromachines 2016, 7, 146 7 of 12 closing time constant can be adjusted. When the air-pressure is stopped, the valve is opened with a500 time constant depending of for the 10 ionic solutions recoveringthe thedifference ionic flow. rpm spin-coating for 30ofs the anddiffusion a 95 °C time baking min. Figure for 10 compares + and Cl− are nearly the same: K+ 64.7 cm/s, Cl− 65.4 cm/s. As KCl is used, the time diffusion for K between commercial Nafion membranes (dry one) where the minimum width you can buy is 40 µm. The selective Nafion membrane is 8 µm wide, located between the two PDMS layers of the The results show that designing the Nafion membrane from Nafion solution significantly improves channels. Nafion membrane is designed with Nafion perfluorinated resin solution 5 wt % using the qualityThe of the in terms of homogeneity, width, and transparency. Micromachines 2016, 7,membrane 146 7 of 12 ◦ C baking for 10 min. Figure 10 compares the difference between 500 rpm spin-coating for 30 s and a 95 Figure 11 shows the whole microfluidic neuron device with the different layers. The size of the commercial Nafionand membranes one) the minimum width you can buy is 40 µm. results different layers described. The PDMS membrane isFigure used for Quake valve design. 500 rpm channels spin-coating for 30are s (dry and a 95where °C baking for 10 min. 10 the compares theThe difference show that designing the Nafion membrane from Nafion solution significantly improves the quality of In this PDMS membrane we create one hole with the where same size the Nafion membrane. between commercial Nafion membranes (dry one) theof minimum width you canThe buyinterface is 40 µm. the membrane in terms of homogeneity, width, and transparency. between theshow two layers channels the center point is separated by the Nafion membrane.improves The results that designing theinNafion membrane from Nafion solution significantly
the quality of the membrane in terms of homogeneity, width, and transparency. Figure 11 shows the whole microfluidic neuron device with the different layers. The size of the different channels and layers are described. The PDMS membrane is used for the Quake valve design. In this PDMS membrane we create one hole with the same size of the Nafion membrane. The interface between the two layers of channels in the center point is separated by the Nafion membrane.
Figure 10. 10. Width Width of of Nafion Nafion membranes Figure membranes computed computedby bylaser laserconfocal confocalmicroscopy. microscopy.InInthe theleft leftpart, part,the the Nafion membrane designed from Nafion solution. In the right part is a 40 µm width dry commercial Nafion membrane designed from Nafion solution. In the right part is a 40 µm width dry commercial Nafion membrane. membrane. The commercial Nafion The homogeneity homogeneity of of the theleft leftpart partisisquite quitebetter betterthan thanthe theright rightpart. part.The The commercial Nafion is also less transparent than the other one. For designed Nafion and commercial Nafion, the the 3 Nafion is also less transparent than the other one. For designed Nafion and commercial Nafion, subfigures represents the width and the surface the membrane. The top left is the computation by 3 subfigures represents the width and the surface the membrane. The top left is the computation by laser confocal confocal microscopy microscopy (bright (bright part part is is glass glass surface, surface, dark dark part part is is membrane membrane surface). surface). The The top top right right is laser Figure 10. Width of Nafion membranes computed by laser confocal microscopy. In the left part, is the topography of membrane the membrane showing the heterogeneity of the membrane. The subfigure bottomthe the topography of the showing the heterogeneity of the membrane. The bottom Nafion membrane designed from Nafion solution. In the right part is a 40 µm width dry commercial subfigure represents width computation of the membrane represents the width the computation of the membrane in µm. in µm. Nafion membrane. The homogeneity of the left part is quite better than the right part. The commercial Nafion is also less transparent than the other one. For designed Nafion and commercial Nafion, the 3 Figure 11 shows the the whole microfluidic neuron withThe thetop different layers. The sizeby of the subfigures represents width and the surface the device membrane. left is the computation different and layers (bright are described. Thesurface, PDMS dark membrane is used forsurface). the Quake design. laserchannels confocal microscopy part is glass part is membrane The valve top right In thisis PDMS membraneofwe one hole with the size of theofNafion membrane. interface the topography thecreate membrane showing thesame heterogeneity the membrane. TheThe bottom between the two layers of in the center point is separated subfigure represents thechannels width computation of the membrane in µm.by the Nafion membrane.
Figure 11. Different layers of the microfluidic neuron.
Figure 12 is composed of three pictures of microfluidic neuron devices. The diameter of the holes is equal to 0.5 mm. The holes are used for filling the four compartments (two for intracellular, two for extracellular regions) with ionic solutions. Two holes of 0.5 mm diameter are used for the air pressure inputs to open/close of the Quake valves. In Figure 12b, the holes are 2 mm in diameter in Figureinfluence 11. Different Different layers of the the microfluidic microfluidic neuron. order to investigate the volume in the compartment. Our results did not reveal any volume Figure 11. layers of neuron. influence. For these three devices, a large Quake valve was selected (500 µm) in order to investigate Figure 12 is composed of three pictures neuron Our devices. The diameter ofthat the holes the size influence and also the diffusion timeofofmicrofluidic the ionic solutions. experiments shows the is equal 0.5 mm. Thethe holes are used for constant filling the compartments (two for intracellular, two larger thetoQuake valve, slower the time of four the repolarizing phase. for extracellular regions) with ionic solutions. Two holes of 0.5 mm diameter are used for the air pressure inputs to open/close of the Quake valves. In Figure 12b, the holes are 2 mm in diameter in order to investigate the volume influence in the compartment. Our results did not reveal any volume
Micromachines 2016, 7, 146
8 of 12
Figure 12 is composed of three pictures of microfluidic neuron devices. The diameter of the holes is equal to 0.5 mm. The holes are used for filling the four compartments (two for intracellular, two for extracellular regions) with ionic solutions. Two holes of 0.5 mm diameter are used for the air pressure inputs to open/close of the Quake valves. In Figure 12b, the holes are 2 mm in diameter in order to investigate the volume influence in the compartment. Our results did not reveal any volume influence. For these three devices, a large Quake valve was selected (500 µm) in order to investigate the size influence and also the diffusion time of the ionic solutions. Our experiments shows that the larger the Micromachines 2016, 7,7,slower 146 88of Quake valve, the the time constant of the repolarizing phase. Micromachines 2016, 146 of12 12
Figure Figure12. 12.333microfluidic microfluidic neurondevices. devices.For Forthe thedevice device(a), (a),the the scale mm. For these three devices, microfluidicneuron thescale scaleisis is222mm. mm.For Forthese thesethree threedevices, devices, the Nafion membrane is 10 µm width. For (a) and (c), holes are 0.5 mm in diameter. For (b), holes the Nafion membrane is 10 µm width. For (a) and (c), holes are 0.5 mm in diameter. For (b), holesare are 22mm for these devices are 100 µm ××10 The Quake valve are mmin indiameter. diameter.The Thefluidic fluidicchannel channel for these devices are 100 µm ×µm. 10 µm. Quake valve are mm in diameter. The fluidic channel for these devices are 100 µm 10 µm. TheThe Quake valve are500 500 µm membrane isis10 width. 500×µm × 50The µm.PDMS The PDMS membrane is 10in µm in width. µm ×50 50µm. µm. The PDMS membrane 10µm µm in width.
4. 4.Results Results Results We generated an action potential with microfluidic neuron device (Figure 13). The We have havegenerated generatedan anaction action potential with our microfluidic neuron device (Figure 13).three The have potential with ourour microfluidic neuron device (Figure 13). The three action potentials are created by air pulses of 100 ms duration and 100 kPa pressure which, three potentials are created by airofpulses 100 msand duration and 100 kPa pressure which, actionaction potentials are created by air pulses 100 msofduration 100 kPa pressure which, respectively, respectively, close the valve 14.1, 16.3 s.s.The has slower respectively, close theQuake Quake valve at 14.1, 16.3 and 18.4 Therepolarizing repolarizing phase has slowervelocity velocity close the Quake valve at 14.1, 16.3at and 18.4 s.and The18.4 repolarizing phase hasphase slower velocity than the than the rising as in than the rising phase aswe weobserve observe inthe thebiological biological signals. rising phase asphase we observe in the biological signals. signals.
Figure 13. In blue, action potentials of the microfluidic neuron; in red, air pressure pulses. action Figure pulses.The The Figure 13. 13. In In blue, blue, action action potentials potentials of of the the microfluidic microfluidic neuron; neuron; in in red, red, air air pressure pressure pulses. The action action potential are created by the Quake valve using air pressure: 100 ms duration and 100 kPa pressure. potential potential are are created created by by the the Quake Quake valve valve using using air air pressure: pressure: 100 100 ms ms duration duration and and 100 100 kPa kPa pressure. pressure. Pulses are created at 14.1, 16.3 and 18.4 s.s.Action potential has an average amplitude of 75 mV, similar Pulses Pulses are are created created at at 14.1, 14.1, 16.3 16.3 and and 18.4 18.4 s. Action Action potential potential has has an an average average amplitude amplitude of of 75 75 mV, mV, similar similar to biological neuron amplitude. to biological neuron amplitude. to biological neuron amplitude.
The The main main parameter parameter for for controlling controlling the the shape shape of of the the action action potential potential isis the the difference difference of of ion ion The main parameter for controlling the shape of the action potential is the difference of+ ion concentration between the two chambers. In the human body, the concentration of potassium (K ) + concentration between the two chambers. In the human body, the concentration of potassium (K )isis concentration between the two chambers. In the human body, the concentration of potassium + 140 140mM mMin inintracellular, intracellular,and and55mM mMin inextracellular, extracellular,environments. environments.The Theconcentration concentrationof ofsodium sodium(Na (Na+)) (K+ ) is 140 mM in intracellular, and 5 mM in extracellular, environments. The concentration of isisfrom from55to to15 15mM mMin inintracellular, intracellular,and and145 145mM mMin inextracellular, extracellular,environments. environments.We Wehave haveperformed performed sodium (Na+ ) is from 5 to 15 mM in intracellular, and 145 mM in extracellular, environments. different experiments with concentrations equivalent to human ones. In a second experiment, different experiments with concentrations equivalent to human ones. In a second experiment, decreased decreased concentrations concentrations with with aa similar similar scale scale factor factor (intra (intra and and extracellular extracellular chambers) chambers) were were performed. The ratio between the intra/extracellular concentrations is the most important factor, performed. The ratio between the intra/extracellular concentrations is the most important factor, as as proved provedby byour ourexperiments. experiments. To To conclude, conclude, we we obtain obtain an an electrical electrical behavior behavior similar similar to to the the biological biological neuron. neuron. For For instance, instance,
Micromachines 2016, 7, 146
9 of 12
We have performed different experiments with concentrations equivalent to human ones. In a second experiment, decreased concentrations with a similar scale factor (intra and extracellular chambers) were performed. The ratio between the intra/extracellular concentrations is the most important factor, as proved by our experiments. To conclude, we obtain an electrical behavior similar to the biological neuron. For instance, biological neuron behaviors are shown in this Nature Nanotechnology paper (Figure 3) [35]. The amplitude, waveform, and timing are similar as our experiments. Our microfluidic neuron can also generate some burst-like behavior (Figure 14). In order to get this kind of response, the opening/closure of the Quake valve has to be sequentially programmed by an air pressure controller. In our experiment, the burst frequency is lower than in biological neurons (by a factor of five). This is due to the hardware performance of our air-pressure controller. Micromachines 2016, 7, 146 9 of 12 Micromachines 2016, 7, 146
9 of 12
Figure 14.In In green,burst-like burst-like behavior of the microfluidic neuron; in red, air pressure pulses. The Figure behavior of the microfluidic neuron; in red, pressure pulses.pulses. The action Figure 14. 14. Ingreen, green, burst-like behavior of the microfluidic neuron; in air red, air pressure The action potentials are created by thevalve Quake valve air pressure: 50 ms duration andpressure. 100 kPa potentials are created by the Quake using airusing pressure: 50 ms duration and 100 kPa action potentials are created by the Quake valve using air pressure: 50 ms duration and 100 kPa pressure. Shown are sequences of four pulses started at 0.50, 2.50 and 4.60 s. Shown are sequences of four pulses started at 0.50, 2.50 and 4.60 s. pressure. Shown are sequences of four pulses started at 0.50, 2.50 and 4.60 s.
The amplitude of the action potential could be determined by the air pressure value (Figure 15). The amplitude of the action potential could be determined by the air pressure value (Figure 15). The tuning of microfluidic neuron action potential amplitude is possible, whereas the behavior is The The tuning tuning of of microfluidic microfluidic neuron neuron action action potential potential amplitude amplitude is is possible, possible, whereas whereas the the behavior behavior is is difficult to obtain by a silicon one. The amplitude tuning is an important issue to mimic the biological difficult difficult to to obtain obtain by by aa silicon silicon one. one. The amplitude tuning is an important issue to mimic the biological signals. The minimum pressure is 20 kPa, as it is the limit (20 mV) for detecting the action potential signals. The minimum pressure is 20 kPa, as it is the the limit limit (20 (20 mV) mV) for for detecting detecting the the action action potential potential from the noise. The maximum pressure was 200 kPa for our device, as more pressure cracked our from the noise. The maximum pressure was 200 kPa for our device, as more pressure cracked the noise. cracked our PDMS membrane. This tuning is linear. PDMS PDMS membrane. membrane. This tuning is linear. linear.
Figure 15. The amplitude of the action potential in function of air pressure of the Quake valves. The Figure 15. 15. The amplitude amplitude of of the the action action potential in in function function of of air air pressure pressure of of the the Quake Quake valves. valves. The The Figure amplitude The can be tuned from 20 to 115 potential mV. amplitude can can be be tuned tuned from from 20 20 to to 115 115 mV. mV. amplitude
The frequency of the spiking activity can be modified by the air pressure controller. The frequency of the spiking activity can be modified by the air pressure controller. minimum duration of this controller was 50 ms, consequently the maximum frequency of minimum duration of this controller was 50 ms, consequently the maximum frequency of microfluidic neuron is 20 Hz. This range is similar to biological neurons. microfluidic neuron is 20 Hz. This range is similar to biological neurons. 5. Discussion 5. Discussion
The The our our
Micromachines 2016, 7, 146
10 of 12
The frequency of the spiking activity can be modified by the air pressure controller. The minimum duration of this controller was 50 ms, consequently the maximum frequency of our microfluidic neuron is 20 Hz. This range is similar to biological neurons. 5. Discussion 5.1. Hybrid Experiments In a final step, hybrid experiments in the same chip were designed: neurons culture in one part of the PDMS device and artificial neurons in the other part. This PDMS device is described in Figure 16. Electrodes on glass are used for stimulation and recording of biological neurons. The microfluidic neuron is used stimulating the biological part by the activation or inactivation of the Quake 10 valves. Micromachines 2016,for 7, 146 of 12
Figure system forfor hybrid experiments. In the “in Figure 16. 16.The Thefirst firstPDMS PDMS(polydimethylsiloxane) (polydimethylsiloxane) system hybrid experiments. In left the part, left part, vitro” neurons cultured on a gold microelectrode array (20 electrodes). In the right part, the “in vitro” neurons cultured on a gold microelectrode array (20 electrodes). In the right part, the microfluidic microfluidic neuron neuron described described previously. previously.
This device is ready for the open-loop (microfluidic neuron to biological neuron) experiments. This device is ready for the open-loop (microfluidic neuron to biological neuron) experiments. For the closed-loop (bidirectional communication) experiment, an electrical board for recording and For the closed-loop (bidirectional communication) experiment, an electrical board for recording and detecting the biological spiking activity is needed. detecting the biological spiking activity is needed. 5.2. 5.2. New New Design Design Ideas Ideas A A key key point point of of the the design design of of such such aa microfluidic microfluidic neuron neuron is is the the membrane membrane exchange exchange between between the the intracellular and extracellular environments. A membrane made of Nafion provides a selectivity intracellular and extracellular environments. A membrane made of Nafion provides a selectivity with with + respect only cations cations (K (K++ and respect to to the the charge charge of of the the ions ions in in passing passing only and Na Na+in inour ourcase). case).Selectivity Selectivityfor forspecific specific ions ions is is not not possible possible with with this this membrane. membrane. The The integration integration of of lipid lipid bi-layers bi-layers [36] [36] would would alleviate alleviate this this problem, problem, and and constitute constitute an an important important method method to to investigate. investigate. The The miniaturization miniaturization of of the the neuron neuron to to approach approach the the biological biological dimensions dimensions (of (of the the order order of of tens tens of of microns), represents an additional step to the technological constraints, particularly on the microns), represents an additional step to the technological constraints, particularly on the integration integration of the selective membrane. of the selective membrane. The neurons should begin before the phase of miniaturization. This The networking networkingofofmicrofluidic microfluidic neurons should begin before the phase of miniaturization. will adapt to any architectural changes required to obtain neural networks with dimensions similar This will adapt to any architectural changes required to obtain neural networks with dimensions to thoseto ofthose biological neurons.neurons. similar of biological A neurons can can be designed in theinnear The plasticity of material A matrix matrixofofmicrofluidic microfluidic neurons be designed thefuture. near future. The plasticity of integration is currently under investigation in order to add synapses between the different artificial material integration is currently under investigation in order to add synapses between the different neurons. artificial neurons. A better miniaturization of the Quake valve can be accomplished. Bursting behavior generated A better miniaturization of the Quake valve can be accomplished. Bursting behavior generated by the microfluidic neuron will be more accurate and with higher frequency. The use of a by the microfluidic neuron will be more accurate and with higher frequency. The use of a piezoelectric piezoelectric valve could be a direction to explore. valve could be a direction to explore. 5.3. Microfluidic Neuron as a Stimulator Another potential approach will be to use the microfluidic neuron as a stimulator for biological neurons. Given that the action potential of the microfluidic neuron has the same characteristics than biological ones, and the ionic concentrations in the microfluidic neuron and in the neuron culture medium are similar, we foresee that the extracellular chamber of the microfluidic neuron and the
Micromachines 2016, 7, 146
11 of 12
5.3. Microfluidic Neuron as a Stimulator Another potential approach will be to use the microfluidic neuron as a stimulator for biological neurons. Given that the action potential of the microfluidic neuron has the same characteristics than biological ones, and the ionic concentrations in the microfluidic neuron and in the neuron culture medium are similar, we foresee that the extracellular chamber of the microfluidic neuron and the neuron culture system can be merged. The microfluidic neuron as a stimulator is, in our opinion, one of the most promising methods to explore in neuromorphic engineering. Acknowledgments: EUJO-LIMMS (Europe-Japan Opening of LIMMS) activities are also supported by the Core-to-Core Program of the Japan Society for the Promotion of Science (JSPS) as a matching fund to the 7th Framework Programme of the European Union. We would like to thank Vincent Raimbault from IMS (Integration du Matériau au Système) Lab., University of Bordeaux, France and Ayako Araki from University of Tokyo, Japan for their support in microfluidic techniques. We thank also Jacques Fattaccioli from ENS (Ecole Normale Supérieure) Paris, France for his support in the design of the last microfluidic neurons. Author Contributions: T.L. and T.F. conceived and designed the experiments; T.L. performed the experiments, analyzed the data and wrote the paper. Conflicts of Interest: The authors declare no conflict of interest.
References 1. 2.
3.
4. 5. 6. 7.
8. 9. 10.
11. 12. 13. 14. 15.
Nicolelis, M.A.L.; Lebedev, M.A. Principles of neural ensemble physiology underlying the operation of brain-machine interfaces. Nat. Rev. Neurosci. 2009, 10, 530–540. [CrossRef] [PubMed] Hochberg, L.R.; Serruya, M.D.; Friehs, G.M.; Mukand, J.A.; Saleh, M.; Caplan, A.H.; Branner, A.; Chen, D.; Penn, R.D.; Donoghue, J.P. Neuronal ensemble control of prosthetic devices by a human with tetraplegia. Nature 2006, 442, 164–171. [CrossRef] [PubMed] Hochberg, L.R.; Bacher, D.; Jarosiewicz, B.; Masse, N.Y.; Simeral, J.D.; Vogel, J.; Haddadin, S.; Liu, J.; Cash, S.S.; Smagt, P.V.D.; et al. Reach and grasp by people with tetraplegia using a neurally controlled robotic arm. Nat. Methods 2012, 485, 372–375. [CrossRef] [PubMed] Mahowald, M.; Douglas, R. A silicon neuron. Nature 1991, 354, 515–518. [CrossRef] [PubMed] Le Masson, G.; Renaud-Le Masson, S.; Debay, D.; Bal, T. Feedback inhibition controls spike transfer in hybrid thalamic circuits. Nature 2002, 417, 854–858. [CrossRef] [PubMed] Ambroise, M.; Levi, T.; Joucla, S.; Yvert, B.; Saighi, S. Real-time biomimetic central pattern generators into FPGA for hybrid experiments. Front. Neurosci. 2013, 7, 215. [CrossRef] [PubMed] Minev, I.R.; Musienko, P.; Hirsch, A.; Barraud, Q.; Wenger, N.; Moraud, E.M.; Gandar, J.; Capogrosso, M.; Milekovic, T.; Asboth, L.; et al. Electronic dura mater for long-term multimodal neural interfaces. Science 2015, 347, 159–163. [CrossRef] [PubMed] Grassia, F.; Buhry, L.; Levi, T.; Tomas, J.; Destexhe, A.; Saighi, S. Tunable neuromimetic integrated system for emulating cortical neuron models. Front. Neurosci. 2011, 5, 134. [CrossRef] [PubMed] Xia, Y.; Whiteside, G. Soft lithography. Annu. Rev. Mater. Sci. 1998, 28, 153–184. [CrossRef] Indiveri, G.; Linares-Barranco, B.; Hamilton, T.J.; Schaik, A.V.; Etienne-Cummings, R.; Delbruck, T.; Liu, S.C.; Dudek, P.; Häfliger, P.; Renaud, S.; et al. Neuromorphic silicon neuron circuits. Front. Neurosci. 2001, 5, 73. [CrossRef] [PubMed] Mead, C. Neuromorphic electronic systems. Proc. IEEE 1990, 78, 1629–1636. [CrossRef] Hodgkin, A.; Huxley, A. A quantitative description of membrane current and its application to conduction and excitation in nerve. J. Physiol. 1952, 117, 500–544. [CrossRef] [PubMed] Indiveri, G. Synaptic plasticity and spike-based computation in VLSI networks of integrate-and-fire neurons. Neural Inf. Process. Lett. Rev. 2007, 11, 135–146. Linares-Barranco, B.; Sénchez-Sinencio, E.; Rodríguez-Vázquez, A.; Huertas, J.L. A CMOS implementation of Fitzhugh-Nagumo neuron model. IEEE J. Solid-State Circuits 1991, 26, 956–965. [CrossRef] Rice, K.L.; Bhuiyan, M.A.; Taha, T.M.; Vutsinas, C.N. FPGA implementation of izhikevich spiking neural networks for character recognition. In Proceedings of 2009 International Conference on Reconfigurable Computing and FPGAs, Cancún, Mexico, 2009; pp. 451–456.
Micromachines 2016, 7, 146
16.
17. 18. 19.
20. 21. 22. 23.
24.
25.
26.
27. 28. 29. 30. 31. 32. 33. 34. 35.
36.
12 of 12
Schemmel, J.; Brüderle, D.; Grübl, A.; Hock, M.; Meier, K.; Millner, S. A wafer-scale neuromorphic hardware system for large-scale neural modeling. In Proceedings of 2010 IEEE International Symposium on Circuits and Systems, Paris, France, 30 May–2 June 2010; pp. 1947–1950. Levi, T.; Lewis, N.; Saïghi, S.; Tomas, J.; Bornat, Y.; Renaud, S. Neuromimetic integrated circuits; Artech House: Norwood, MA, USA, 2008; pp. 241–264. Cassidy, A.; Georgiou, J.; Andreou, A.G. Design of silicon brains in the nano-CMOS era: Spiking neurons, learning synapses and neural architecture optimization. Neural Netw. 2013, 45, 4–26. [CrossRef] [PubMed] Simoni, M.F.; Cymbalyuk, G.S.; Sorensen, M.E.; Calabrese, R.L.; DeWeerth, S.P. A multi-conductance silicon neuron with biologically matched conductances. IEEE Trans. Biomed. Eng. 2004, 51, 342–354. [CrossRef] [PubMed] Simoni, M.F.; DeWeerth, S.P. Sensory feedback in a half-center oscillator model. IEEE Trans. Biomed. Eng. 2007, 54, 193–204. [CrossRef] [PubMed] Kohno, T.; Sekikawa, M.; Li, J.; Nanami, T.; Aihara, K. Qualitative-modeling-based silicon neurons and their networks. Front. Neurosci. 2016, 10, 273. [CrossRef] [PubMed] Kohno, T.; Aihara, K. Mathematical-model-based design method of silicon burst neurons. Neurocomputing 2008, 71, 1619–1628. [CrossRef] Joucla, S.; Ambroise, M.; Levi, T.; Lafon, T.; Chauvet, P.; Saïghi, S.; Bornat, Y.; Lewis, N.; Renaud, S.; Yvert, B. Generation of locomotor-like activity in the isolated rat spinal cord using intraspinal electrical microstimulation driven by a digital neuromorphic CPG. Front. Neurosci. 2016, 10, 67. [CrossRef] [PubMed] Bonifazi, P.; Difato, F.; Massobrio, P.; Breschi, G.L.; Pasquale, V.; Levi, T.; Goldin, M.; Bornat, Y.; Tedesco, M.; Bisio, M.; et al. In vitro large-scale experimental and theoretical studies for the realization of bi-directional brain-prostheses. Front. Neural Circuits 2013, 7, 40. [CrossRef] [PubMed] Biffi, E.; Piraino, F.; Pedrocchi, A.; Fiore, G.B.; Ferrigno, G.; Redaelli, A.; Menegon, A.; Rasponi, M. A microfluidic platform for controlled biochemical stimulation of twin neuronal networks. Biomicrofluidics 2012, 6, 24106–24110. [CrossRef] [PubMed] Biffi, E.; Menegon, A.; Piraino, F.; Pedrocchi, A.; Fiore, G.B.; Rasponi, M. Validation of long-term primary neuronal cultures and network activity through the integration of reversibly bonded microbioreactors and MEA substrates. Biotechnol. Bioeng. 2012, 109, 166–175. [CrossRef] [PubMed] Popovic, M.A.; Carnevale, N.; Rozsa, B.; Zecevic, D. Electrical behaviour of dendritic spines as revealed by voltage imaging. Nat. Commun. 2015, 6, 8436. [CrossRef] [PubMed] Potter, S.M. Closing the loop between neurons and neurotechnology. Front. Neurosci. 2010, 4, 15. [CrossRef] [PubMed] Gross, P.; Kartalov, E.; Scherer, A.; Weiner, L. Applications of microfluidics for neuronal studies. J. Neurol. Sci. 2007, 252, 135–143. [CrossRef] [PubMed] Taylor, A.; Blurton-Jones, M.; Rhee, S.W.; Cribbs, D.; Cotman, C.; Jeon, N.L. A microfluidic culture platform for CNS axonal injury, regeneration and transport. Nat. Methods 2005, 2, 599–605. [CrossRef] [PubMed] Park, J.W.; Vahidi, B.; Taylor, A.; Rhee, S.W.; Jeon, N.L. Microfluidic culture platform for neuroscience research. Nat. Protoc. 2006, 1, 2128–2136. [CrossRef] [PubMed] Unger, M.; Chou, H.-P.; Thorsen, T.; Scherer, A.; Quake, S. Monolithic microfabricated valves and pumps by multilayer soft lithography. Science 2000, 288, 113–116. [CrossRef] [PubMed] Lee, J.H.; Song, Y.A.; Han, J. Multiplexed proteomic sample preconcentration device using surface-patterned ion-selective membrane. Lab Chip 2008, 8, 596–601. [CrossRef] [PubMed] Quist, J.; Trietsch, S.; Vultoa, P.; Hankemeier, T. Elastomeric microvalves as tunable nanochannels for concentration polarization. Lab Chip 2013, 13, 4810–1815. [CrossRef] [PubMed] Robinson, J.; Jorgolli, M.; Shalek, A.; Yoon, M.H.; Gertner, R.; Park, H. Vertical nanowire electrode arrays as a scalable platform for intracellular interfacing to neuronal circuits. Nat. Nanotechnol. 2012, 7, 180–184. [CrossRef] [PubMed] Funakoshi, K.; Suzuki, H.; Takeuchi, S. Lipid bilayer formation by contacting monolayers in a microfluidic device for membrane protein analysis. Anal. Chem. 2006, 78, 8169–8174. [CrossRef] [PubMed] © 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).
