{"title": "Electronic Receptors for Tactile/Haptic Sensing", "book": "Advances in Neural Information Processing Systems", "page_first": 785, "page_last": 792, "abstract": null, "full_text": "ELECTRONIC  RECEPTORS  FOR  TACTILE/HAPTIC\u00b7  SENSING \n\n785 \n\nElectrical and Computer Engineering \n\nThe Johns Hopkins University \n\nAndreas G. Andreou \n\nBaltimore, MD 21218 \n\nABSTRACT \n\nWe discuss synthetic receptors for  haptic sensing. These are based on \nmagnetic field sensors (Hall effect structures) fabricated using standard \nCMOS technologies.  These receptors, biased with a small permanent \nmagnet can detect the presence of ferro or ferri-magnetic objects in the \nvicinity of the sensor. They can also detect the magnitude and direction \nof the magnetic field. \n\nINTRODUCTION \n\nThe organizational structure and functioning of the sensory periphery in living beings has \nalways been the subject of extensive research. Studies of the retina and the cochlea have \nrevealed  a  great  deal  of information  as  to  the  ways  information  is  acquired  and \npreprocessed;  see  for  example  review  chapters  in  [Barlow  and  MoHon,  1982]. \nUnderstanding of the principles  underlying  the  operation of sensory  channels  can  be \nutilized to develop machines that can sense their environment  and function  in it, much \nlike living beings. Although vision is the principal sensory channel to the outside world, \nthe  \"skin senses\"  can  in  some cases provide information  that is  not  available  through \nvision.  It is interesting  to  note, that the performance in identifying objects  through  the \nhaptic senses can be comparable to vision  [Klatzky el al,  1985]; longer learning periods \nmay be necessary though. Tactually guided exploration and shape perception for robotic \napplications has been extensively investigated by [Hemanli el al, 1988]. \n\nA number of synthetic sensory  systems  for  vision and audition based on physiological \nmodels for the retina and the cochlea have been prototyped by Mead and his coworkers in \nVLSI  [Mead,  1989].  The key  to  success  in  such  endeavors  is  the ability  to  integrate \ntransducers (such as light sensitive devices) and  local  processing electronics on the same \nis  silicon  CMOS;  furthermore,  it is \nchip.  A  technology  that  offers  that  possibility \nreadily  available  to  engineers  and  scientists  through  the  MOSIS  fabrication \nservices[Cohen and Lewicki, 1981]. \n\nReceptor  cells,  are  structures  in  the  sensory  pathways  whose  purpose  is  to  convert \nenvironmental  signals  into  electrical  activity  (strictly  speaking  this  is  true  for \n\n\u2022  Haptic refers to the perception of vibration, skeletal conformation or position and \nskin deformation. Tactile refers to the perceptual system that includes only the \ncutaneous senses of vibration and deformation. \n\n\f786 \n\nAndreou \n\nexteroceptors). The retina rods and cones are examples of receptors for light stimuli and \nthe Pacinian corpuscles are mechanoreceptors that are sensitive to indentation or pressure \non the skin.  A synthetic receptor is thus the first and necessary functional element in any \nsynthetic sensory system. For the development of vision systems parasitic bipolar devices \ncan be used [Mead, 1985]  to perform the necessary light to electrical signal transduction \nas well as low level signal amplification. On the other hand, implementation of synthetic \nreceptors  for  tactile perception  is  still  problematic  [Barth et.  al.,  1986].  Truly  tactile \ntransducers  (devices sensitive to pressure stimuli) are not available in standard CMOS \nprocesses and are only found in specialized fabrication lines. However, devices that are \nsensitive to magnetic fields can be used to some extend as a substitute. \n\nIn this paper, we discuss the development of electronic receptor elements that can be used \nin synthetic haptic/tactile sensing systems. Our receptors are devices which are sensitive \nto steady state or varying magnetic fields  and give electrical signals proportional to the \nmagnetic induction. They can all be fabricated using standard silicon processes such as \nthose offered by MOSIS. We show how our elements can be used for tactile and haptic \nsensing and compare  its characteristics  with  the features  of biological  receptors.  The \nspatial resolution of the devices, its frequency response and dynamic range  are more than \nadequate. One of our devices has nano-watt power diSSipation and thus can be used in large \narrays for high resolution sensing. \n\nTHE  MAGNETIC-FIELD  SENSORY  PARADIGM \n\nIn this  section we show  qualitatively how  to  implement synthetic sensory functions  of \nthe haptic and tactile senses by using magnetic fields  and their interaction with ferri  or \nferro-magnetic objects.  This will motivate the more detailed discussion that follows on \nthe transducer devices. \n\nDIRECT  SENSING: \n\nIn this mode of operation the transducer will detect the magnitude and direction of the \nmagnetic  induction  and convert  it  into  an  electrical  signal.  If the  magnetic  field  is \nprovided by the fringing  fields  of a small permanent magnet, the strength of the signal \nwill falloff with the distance of the sensor from the magnet Such an arrangement for one \ndimensional  sensing is  shown in Figure  1. The experimental  data are from  the  MOS \nHall-voltage  generator  that  is  described  in  the  next  section.  The magnetic  field  was \nprovided  by  a  cylindrical,  rare-earth,  permanent  magnet  with  magnetic  induction \nB0=250mT, measured on  the end surfaces.The vertical axis shows  the signal from  the \ntransducer (Hall-voltage) and the horizontal axis represents  the distance d of the sensor \nfrom  the surface of the magnet. \n\nThe above scheme can be used to sense the angular displacement between two fingers at a \njoint (inset b).  By placing a small magnet on one side of the joint and the receptor on the \nother, the signal from the receptor can be conditioned and converted into a measure of the \nangle e between the two fingers. The output of our receptor would thus correspond to the \noutput  from  the Joint Fibers  that originate in the Joint Capsule  [Johnson,  1981]. Joint \nangle perception and manual stereognosis is mediated in part by these fibers. The above is \njust one  example of how to use our integrated electronic receptor element for  sensing \n\n\fElectronic Receptors for TactileIHaptic Sensing \n\n787 \n\nskeletal conformation and position. Since there is no moving parts other than  the joint \nitself,  this is a reliable scheme. \n\nSensor \n\nI 1 MAGNET~ \n\nV \n\n(8) \n\n2.63 \n\n2.~ \n\n1.88 \n\n1.50 \n\n1.13 \n\n.750 \n\n.315 \n\nMagnet \n\nSensor \n\n(b) \n\nO.OO~--~~--~----r---~----,-----~ __ ~ ____ , -__ ~~ __ ~ \n10. \n\n9.00 \n\n6.00 \n\n7.00 \n\n8.00 \n\n3.00 \n\nIf.OO \n\n5.00 \n\n0.00 \n\n1.00 \n\ne.oo \n\nDistance  ot  magnet  trom  sensor  (mm) \n\nFigure 1. Direct Sensing Using an Integrated Hall-Voltage Transducer \n\nPASSIVE  SENSING \n\nIn this mode of operation, the device or an array of devices are permanently biased with a \nuniform  magnetic field whose uniformity can be disturbed in the presence of a ferro or \nferri-magnetic  object  in  the vicinity.  Signals from  an  array of such  elements would \nprovide ifnormation on the shape of the object that causes the disturbance of the magnetic \nfield.  Non-magnetic  objects can be  sensed if the  surface of the chip  is covered  with  a \ncompliant membrane  that has  some  magnetic  properties.  Note  that our receptors  can \ndetect the presence of an  object without having a  direct contact with  the object itself. \nThis may in some cases be advantageous. \n\nIn  this application,  the magnetic field  sensor would act more like the  Ruffini  organs \nwhich exist in  the deeper tissue and are primarily  sensitive to  static  skin  stretch.  The \nabove scheme could also be  used for sensing dynamic. stimuli and there is a  variety of \nreceptor cells. such as the Pacinian  and Meissner's corpuscles  that perfonn that function \nin biological tactile senses [Johnson.  1981]. \n\n\f7BB \n\nAndreou \n\nSILICON  TRANSDUCERS \n\nMagnetic field sensors can be integrated on silicon in a variety of forms. The transduction \nmechanism  is  due  to  some  galvanomagnetic  effect;  the  Hall  effect  or  some  related \nphenomenon  [Andreou.  1986].  For an  up-todate  review  of integrated magnetic  field \nsensors  as  well  as  for  the  fme  points of the  discussion  that  follows  in  the  next  two \nsections. please refer to [Baltes and Popovic. 1986]. The  simplest Hall device is the Hall(cid:173)\nvoltage sensor. This is a four terminal device with two current terminals and two voltage \nterminals  to  measure the Hall-voltage (Figure 2).  A  magnetic field  B  in the direction \nperpendicular to the current flow. sets up the Hall-voltage in the direction indicated in the \nfigure. The Hall-voltage is such that it  compensates for the Lorentz force on the charge \ncarriers. In the experiment below. we have used a MOS Hall generator instead of a bulk \ndevice. The two current contacts are the source and drain of the device and the voltage \ncontacts are formed by small diffusion areas in the channel. The Hall-voltage is linearly \nrelated to  the  magnetic  induction B  and the total current in the sample I  between the \ndrain and the source which is controlled by the gate voltage V gs. \n\n3.00r-------------------------------~--------~--------~--~~~--------~ \n\nM89P  n-t \n\nVds=5V \n\n\" ::> e \n\n'>J' --IU \n\nl: \nI \n:::> \n\n1.25 \n\n.750 \n\nD.OO \n\n-.7:50 \n\n-I.\" \n\n-2.25 \n\nB=250mT \n\nB=-2S0mT \n\n-3 .01  +------r-----~--.,.---_r_----_r------_r_----_r------_,__-_..-~ \n5.1 \n\n1.5 \n\n1.5 \n\n3.0 \n\n2.0 \n\n0.0 \n\n.SO \n\n1.0 \n\n3.' \n\n\".0 \n\n't.5 \n\nVgs  (V) \n\nFigure 2. An Integrated MOS Hall-Voltage Sensor and Measured Characteristics \n\n\fElectronic Receptors for TactileIHaptic Sensing \n\n789 \n\nThe  constant  of proportionality  K  is  related  to the dimensions  of the device and to \nsilicon  electronic transport parameters. The device dimensions and biasing conditions are \nshown in the figure above. Note that the Hall voltage reverses. when the direction of the \nmagnetic field is reversed. The above device was fabricated in a standard 2-micron n-well \nCMOS  process  through  MOSIS  (production  run  M89P).  The  signal  output  of this \nsensor is a  voltage and is relatively  small.  For the same biasing conditions. the signals \ncan be increased only if the channel is made shorter (increase the transconductance of the \ndevice). On the otherhand. when the length of the device approaches its width the Hall(cid:173)\nvoltage  is  shorted  out by  the  heavily doped  source and  drain  regions  and  the  signal \ndegrades again. Some of the problems with the Hall-voltage sensor can be avoided if we \nuse a device that gives a current as its output signal; this is discussed in the next section. \n\nTHE  HALL-CURRENT\u00b7  SENSOR \n\nlI-t \n\nVgs \n\n0 \n\n0 \n\nB=  250mT \n\n\" <t \n\"\" ..... \n..... \nf11 \n::r: \nI \n1-4 \n\n.1-11 \n\n10-11 '!::--'-!---,---,-----r----y----,.---r-----,r----r----I \n1.0 \n\n.7a \n\n.65 \n\n.50 \n\n.55 \n\n.85 \n\n.90 \n\n.60 \n\n.75 \n\n.80 \n\n.95 \n\nFigure 3. The Hall-Current Sensor \n\n\u2022  Hall current is a misnomer (used also by this author) that exists in the literature  to \ncharacterize the current flow in the sample when the Hall  field is shorted out. \n\n\f790 \n\nAndreou \n\nThis  current is a  direct consequence  of the Lorentz  force  and  is perpendicular to  the \ndirection of cmrent flow without a magnetic field The Lorentz force: \n\nt= qO  X e \n\ndepends on the velocity of the carriers in the sample and on the magnetic induction. Since \nthis force is responsible for the transverse cmrent flow a more appropriate name for this \nsensor is a Lorentz-current  sensor. Obviously. given a  magnetic  field  strength. if we \nwant a  maximum  signal  from  our device we  want the carriers  to  have the maximum \nvelocity in the channel. We achieve that by operating the devices in the saturation region \nwere the carriers transverse the channel at what is called the \"saturation velocity\". In this \nconfiguration we can also use  short channel devices (and consequently smaller devices) so \nthat the high fields in the channel can be set with lower voltages. The geometry for such \nsensor as  well  as  the  biasing  circuit  is  shown  in  the  insets  of Figure 3.  As  with the \nprevious sensor. the magnetic field is applied in the direction perpendicular to the channel \n(also perpendicular to the plane of the gate electrode). \n\nThis sensor is a MOS  transistor  with  three terminals and the gate. It has a single source \nbut a split drain.  The poly silicon gate is extended in the region between the two drains so \nthat they are electrically independent of each other. The device is biased with a constant \ndrain-source voltage (with the two drains at the same potential) and the cmrents from the \ntwo sources are monitored. The current-mode circuitry for the synthetic neurons described \nby K wabena [K wabena et al .\u2022  1989] can be employed for this function. We operate the \ndevice in the subthreshold region (denoted by the gate-source volrages between 0.5 and 0.9 \nvolts).  On  the  application  of a  transverse  magnetic  field  we  observe  the  imbalance \nbetween  the  two  drain  currents.  The  Hall-current.  plotted  in  Figure  3.  is  twice  that \ncmrent \n\nNote that we can operate the device and observe an effect due to the magnetic field at very \nlow currents in the nano-amp range. The graph in Figure 3 shows also the dependence of \nthe Hall-current on the gate voltage. This is a logarithmic relationship because the Hall(cid:173)\ncurrent  is  directly  related  to  the  total  current  in  the  sample  Ids  through  a  linear \nrelationship;  it  is  also  linearly  related  to  the  magnetic  field  with  a  proportionality \nconstant Kh. The derivation of a fonnula for the Hall-cmrent can be found in [Andreou. \n1986]. \n\nDISCUSSION \n\nThe frequency  response of the sensors described above is  more than adequate for  our \napplications.  The frequency  response of the Hall  effect mechanism  is many  order of \nmagnitude higher than the requirements of our circuits which have to work only the Hz an \nkHz range.  Another important criterion for  the receptors is their spatial resolution. We \nhave fabricated and tested devices with areas as small as  144 square microns. These are \ncomparable or even  better with what is found in biological systems (10 to 150 receptors \nper square cm). On the otherhand. it is more likely that our final  \"receptor\" elements will \nbe larger. partly. because of the  additional electronics.  The experimental  data shown \nabove  are  only  for  stimuli  that are  static  and  are  simply  the  raw  output  from  our \n\n\fElectronic Receptors for Tactile/Haptic Sensing \n\n791 \n\ntransducer itself.  Clearly such output signals will not be of much use without some local \nprocessing. For example, adaptation mechanisms  have to be included in our synthetic \nreceptors for cutaneous  sensing. The sensitivity  of our transducer elements may be  a \nproblem.  In  that  case  more  sophisticated  structures  such  as  parasitic  bipolar \nmagnetotransistors or combination of MOS Hall and bipolar devices can be employed for \nlow level signal amplification [Andreou, 1986]. Voltage offsets in the Hall-voltage sensor \nwould  also  present  some  problem;  the  same  is  true  for . current  imbalances  due  to \nfabrication imperfections in the Hall current sensor. One of the most attractive properties \nof the  Hall-current type sensor described in  this  paper is its ability to work  with  very \nlower voltages and very low currents; one  of our devices can operate with bias voltage as \nlow  as  3SOmV  and  total cmrent of InA without compromising  its sensitivity.  Power \ndissipation may be a problem  when large arrays of these devices are considered. \n\nDevices for sensing temperature can also be implemented on a standard silicon CMOS \nprocess. Thus a multisensor chip, could be designed that would respond to more than one \nof the somatic senses. \n\nCONCLUSIONS \n\nWe have demonstrated how to use the magnetic field as  a paradigm  for  haptic sensing. \nWe  have  also reported  on  a  silicon  magnetic  field  sensor  that  operates  with  power \ndissipation as low as 3SOpW without any compromise in its performance. This is a dual(cid:173)\ndrain MaS Hall-current device operating in  the subthreshold region.  Our elements are \nonly very simple \"receptors\" without any adaptation mechanisms or local processing; this \nwi11 be the next step in our work. \n\nAcknowledg  ments \n\nThis work was funded by the  Independent Research  and Development program of the \nJohns Hopkins University, Applied Physics Laboratory. The support and personal interest \nof Robert Jenkins is gratefully acknowledged. The author  has benefited by the occasional \ncontact with  Ken  Johnson of the Biomedical Engineering Department \n\nReferences \n\nH.B.  Barlow and J.D.  Mollon  eds.,The  senses, Cambridge University  Press,  Oxford, \n\n1982. \n\nRL. Klatzky, S1. Lederman and V.A. Metzger, \"Identifying objects by touch:  An 'expert \n\nsystem',\" Percept. Psychophys. vol. 37. 1985. \n\nH. Hemami, I.S. Bay and R.E. Goddard, \"A Conceptual Framework for Tactually Guided \nExploration and Shape Perception,\" IEEE Trans. Biomedical Engineering, vol.  35, No. \n2,  Feb.  1988. \n\nC.A. Mead. Analog VLSI and Neural Systems, Addison and Wesley, (in press). \nD.  Cohen  and G.  Lewicki,  \"MOSIS-The  ARPA  silicon  broker,\"  Proc.  of the  Second \n\nCaltech Conference on VLSI,  Pasadena, Califomia,1981. \n\nC.A. Mead,  A sensitive electronic photoreptor, 1985 Chapel Hill Conference  on VLSI, \n\nChapel Hill,  1985. \n\n\f792 \n\nAndreou \n\nP.w. Barth. M.1.  Zdeblick. Z.  Kuc  and P.A.  Beck. \"Flexible tactile arrays for robotics: \narchitectural  robustness  and  yield  considerations.  Tech.  Digest.  IEEE  Solid  State \nSensors Workshop. Hilton Head Island. 1986. \n\nA.G. Andreou. \"The Hall effect and related phenomena in microelectronic devices.\" Ph.D. \n\nDissertation. The Johns Hopkins University. Baltimore. MD.  1986. \n\nH.P.  Baltes  and  R.S.  Popovic.  \"Integrated  Semiconductor  Magnetic  Field  Sensors.\" \n\nProceedings IEEE.  vo1.74.  No.8. Aug.  1986. \n\nK.O.  Johnson  and  G.D.  Lamb,  \"Neural mechanisms  of spatial discrimination:  Neural \npatterns evoked by  Braille-like dot patterns in the monkey.\" 1. of Phys .\u2022  vol.  310. pp. \n117-144. 1981. \n\nK.A.  Boahen.  A.G.  Andreou.  P.O.  Pouliquen  and  A.  Pavasovic.  \"Architectures  for \nAssociative Memories Using Current-Mode Analog MOS Circuits:' Proceedings of the \nDecennial Caltech Conference on VLSI. C. Seitz ed. MIT Press. 1989. \n\n\fAppendix \n\nSummaries of Invited Talks \n\n\f", "award": [], "sourceid": 128, "authors": [{"given_name": "Andreas", "family_name": "Andreou", "institution": null}]}