IMIC - Needle-Shaped Low-Power MonolithicActive Pixel Sensor for Molecular Neuroimaging on

Awake and Freely Moving Rats

Julian Heymes∗, Luis Ammour†, Matthieu Bautista‡, Gregory Bertolone∗, Andrei Dorokhov∗, Sylvain Fieux§,Fabrice Gensolen‡, Matthieu Goffe∗, Fadoua Guezzi-Messaoud∗, Christine Hu-guo∗, Maciej Kachel∗,

Francoise Lefebvre†, Frederic Pain†, Patrick Pangaud‡, Laurent Pinot†, Marc Winter∗, Pascale Gisquet-Verrier¶,Philippe Laniece†, Christian Morel‡, Marc-Antoine Verdier†, Luc Zimmer§ and Jerome Baudot∗

∗Universite de Strasbourg, CNRS, IPHC UMR 7178, F-67000 Strasbourg, FranceEmail: julian.heymes@iphc.cnrs.fr

†IMNC UMR 8165, Universite Paris-Sud, Universite Paris Diderot, CNRS/IN2P3, Universite Paris-Saclay, 91405 Orsay, France‡Aix Marseille Universit, CNRS/IN2P3, CPPM UMR 7346, 13288, Marseille, France

§LNRC, CNRS/INSERM, Univ. Lyon 1, Lyon, France¶NeuroPSI, CNRS/INSB, Univ. Paris Sud, Orsay, France

Abstract—IMIC is a Monolithic Active Pixel Sensor prototypefor the MAPSSIC project dedicated to direct detection of lowenergy β+ rays in the brain of awake and freely-moving ratsusing CMOS technology. Former experiments using a β+ Siprobe developed within the PIXSIC project validated a method-ological proof of concept. However, conducting routinely suchmeasurements would require improvements with respect to thepassive pixel sensors employed in PIXSIC. The new IMIC circuitis fabricated in a 180 nm CMOS Image Sensor Technology andfeatures a matrix of 16 x 128 pixels, which are 30 x 50 µm2

large. The sensor has a needle-like aspect ratio (610 µm x 12000 µm). The chip is produced on a 18 µm high-resistivityepitaxial layer substrate. The foreseen application requires highsensitivity to β-rays while being immune to background γ-rays.Another severe constraint is the limited power dissipation inorder to minimize the thermal impact on the brain. IMIC isa fully-programmable digital sensor. The pixel design is basedon the front-end architecture of the ALPIDE chip. Howevermodifications have been made to store the information insidefired pixels between two readouts allowing low data throughput.The circuit is controlled through the SPI protocol, which allowsfor setting all the necessary polarization signals. The resultsof post-layout simulations show a high signal to noise ratio(>40) and low power dissipation of 115 µW/matrix. Laboratorycharacterization using β-rays validate these predictions anddemonstrated that the slow readout can cope with the expectedlow activity (≈ 120 hits/matrix/s).

Index Terms—sensor, in-vivo molecular imaging, β+, MAPS

I. INTRODUCTION

Recent studies pointed out the need of PET neuroimagingon behavioral studies [1] with awake and freely moving ratsto remove the contraints and anaesthesics biasing. Severalbehavioral neuroimaging methods based on rat PET have beenstudied for the past ten years [2]

• The simplest approach consists of conducting behavioralstudies on a freely moving rodent after the injection of

a radiotracer and before being anesthezied to acquirebrain images with a µPET [3]. The neuroimaging andthe behavioural studies are not conducted simultaneously,thus limiting the range of the possible applications.

• The RatCAP [4], a wearable PET device capable ofperforming full brain images on awake and freely movingrats. Because of the size and the weight of this device, ithas to be held by an articulated arm, which therefore isrestraining the rodent.

• Motion tracking in a PET scanner where the rodent isrestrained in a small area [5], [6], [7]. It also requirespost measurements analysis for reconstruction.

Another approach consists of invasive measurements witha probe implanted by stereotaxic surgery in specific regions.Two previous versions have been developed:

• SIC (Sonde Intra Cranienne) is composed of either aplastic scintillator developed at IMNC [8] and at the PETcenter of University Hospital of Zurich (β-probe)[9] ora LSO crystal developed at the BNL (NY, USA)[10].The probe is coupled to an optical fiber read out with aphotomultiplier tube operated in a single photon countingmode. The connection ensured by the optical fiber is stillrestraining the behavior of the rodent.

• PIXSIC (PIXelated Sonde Intra Cranienne) [11], [12]is the first attempt of using a pixelated sensor in thebrain. The setup is illustrated in Fig. 1. It permits lessrestraining and free movements of the rat by embeddingthe power and control units in a backpack with real-time data acquisition through wireless transmission. ThePIXSIC system is composed of a Silicon probe of 10passive pixels of 200 x 500 µm2. The probe is thinned

down to 200 µm but still exhibits too high sensitivityto the 511 keV γ-Rays while becoming quite brittle.Also, the passive probe is coupled to its read-out ASICcalled PICPUS. The length of the metal lines betweenthe sensing nodes and the first preamplifier stage is2 cm long inducing a problematic sensitivity to externalperturbances. Thus the SNR of the sensor is limited buthigh enough to validate the pharmacological suitability[13], [14].

Fig. 1. Schematic representation of the PIXSIC system. A passive pixelatedprobe with its read-out ASIC (PICPUS) connected to a microcontroller withwireless data transmission.

Taking into account these limitations, we have decided toinvestigate the potential of Monolithic Active Pixel Sensor(MAPS) to obtain an improved solution of our detector withregards to our application. We describe in this paper a firstversion of the MAPS sensor, called IMIC. Preliminary mea-surements are presented in section III. Section IV. concludesand discusses the perspectives of this work.

II. IMIC SENSOR DESIGN

The limitations of passive pixelated sensors make MAPSattractive for low energy β+ counting. The amplifying stagebeing embedded within the pixel mitigates strongly the noiseissue and increases the sensitivity to low energy β+. MAPSalso features thin sensitive volume of only a few tens ofmicrons which virtually suppresses the γ-rays background.Last but not least, the advantage of monolithic sensors isin-chip data processing, thus avoiding the necessity of anadditional read-out ASIC and allowing direct connection to amicrocontroller as depicted in Fig. 2. The system being morecompact is expected to be mechanically more robust.Nevertheless, active pixel sensors dissipate heat that mightalter the brain functions. Special care has to be taken todesign a sensor with a minimal power disspation. It isdifficult to derive an absolute value for the maximal powerthat can be allowed. To obtain an order of magnitude,we performed a naive computation taking into accountthe volume and the thermal capacity of the brain. For atypical experiment duration of 2 hours, allowing a maximumaverage temperature rise of 0.5 C leads to a limit of 500 µW .

A new sensor design is proposed for the MAPSSIC project:IMIC (Imageur Moleculaire Intra Cerebral). The size andshape of the sensor are chosen to minimize its invasiveness into

Fig. 2. Schematic of the MAPSSIC system: the IMIC sensor is directly con-nected to the micro-controller without a read-out ASIC. IMIC is 12000 µmlong and 610 µm wide.

the brain while staying robust enough to be implanted. Thus,the MAPS is shaped like a needle: 610 µm wide, 12000 µmlong, and 200 µm thick. The sensor is designed in 0.18 µmCMOS image sensor technology, and processed on a highresistivity (≥ 1 kΩ · cm) 18 µm thick sensitive epitaxiallayer, which will filter the 511 keV γ-rays background.The sensitive area is located on one end of the sensor andis composed of 16 x 128 pixels of 30 x 50 µm2. Theactive area represents 42 % of the whole sensor. The front-end architecture is based on the one of the ALPIDE sensor(ALice PIxel DEtector) [15]. It was chosen for its low powerdissipation (55 nW/pixel) and high SNR. Whenever a signalappears on the collecting diode, it is amplified, shaped, anddisciminated. If the discriminator output is positive, it indicatesthat the pixel is fired. A 1-bit memory then stores the results ofthe discrimination to be read out in rolling-shutter sequence.The pixel architecture containing the amplifier and the 1-bitmemory is represented in Fig. 3. The front-end architecture isoperating without dead time but although pile-up may occurfor long integration time.

Fig. 3. Schematic of the in-pixel amplifier of IMIC. The amplifier part isbased on the front-end architecture of ALPIDE [15] to which one adds a 1-bitmemory in order to store the information of the hit between two read-outs.

During the rolling shutter readout, the data of each columnin a row is serialized to the output at fixed frequency f thatdetermines the peaking bandwidth. The time needed to read

out the whole matrix is:

treadout = 2048 · f

The foreseen activity derived from PIXSIC experiments [13],[14] is ∼ 100 hits/matrix/s. In order to limit the overalldata rate handled by the micro-controller, readout sequencesare separated by a waiting time defined by the user to keepthe sensor below the saturation, depending on the applicationand activity of the radio-tracer. The whole operation of IMIC,depicted in Fig. 4, determines the integration time:

tintegration = treadout + twaiting

and the final data rate:

Data Rate =2048

tintegration[bits/s]

The digital power dissipation to perform the operation of thesensor varies with the integration time. It ranges between1 and 3 µW .

Fig. 4. Readout operation with IMIC. The readout time is fixed. The waitingtime is set by the needs of the experiments. The integration time is thecombination of the readout time and the waiting time. When a pixel is fired,the information of the hit is stored in the in-pixel memory until the nextreadout.

The pads are located on the sensor end opposite to thematrix. In order to reduce the connexions needed between thesensor and the microcontroller, a SPI protocol is implementedto steer on-chip DACs, which set the polarization of the front-end transistors. The SPI command decoder and the DACsare located on the chip side staying outside the brain, or atleast close to the interface in order to minimize the impact ofheating caused by the commutations of the digital part.

III. SENSOR VALIDATION

Several measurements were taken in order to validate theIMIC sensor.The measured power consumption for the whole sensor is161 µW . Also, on-chip DACs are fully operational aftersetup using the SPI protocol and can polarize the front-end.Measurements have been taken using a 90Sr β source withvarious aluminum shield thicknesses (1, 3, and 7 mm) toregulate the effective activity on the chip. The thickest shieldwas chosen to reproduce the activity expected in the final

experiment.

For each thickness, 1000 frames were taken to acquire themean entries for various integration times between 10 ms and1 s. A similar measurement was performed without sourceto obtain the dark count rate (DCR). The mean entries perintegration time were converted into a mean activity over 1000frames:

Mean Activity(tintegration) =entries

tintegration[hits/matrix/s]

The sensor is saturated when, for any integration time everypixel was touched. It defines the maximum activity reachablefor a given integration time.

saturation(tintegration) =2048

tintegration[hits/matrix/s]

The average activities measured in the various conditions aswell as the saturation limit are presented in Fig. 5.

Fig. 5. Mean activity measured over 1000 frames with the IMIC sensor versusintegration time. The highest activity follows the saturation limit of the sensor.For activities below the limit, no hits are lost for any integration time up to 1s.For long integration time, the DCR is ∼ 1.15. The maximum DCR is observedfor integration time around 50 ms measured at about 3 hits/matrix/s

• The DCR exhibits a soft dependence with the integra-tion time, decreasing from about 3 hits/matrix/s fortintegration ≤ 50 ms to 1.5 hits/matrix/s for thelongest integration time of 1 s. This level of DCR stayscompatible with a robust measurement for activities downto 10 hits/matrix/s.

• With the thinnest shielding, the sensor is saturated forintegration times above 20 ms. The maximum activityseen by the sensor is ∼ 80000 hits/matrix/s.

• At activities below the saturation limit(≤ 2048 hits/matrix/s) obtained with the shieldingsof 3 and 7 mm, the mean activity remains the same forany integration time showing lossless acquisition withlow overall data rate.

IV. CONCLUSION AND PERSPECTIVES

A fully operational full-scale needle-shaped CMOS activepixel sensor has been designed for the purpose of intracranialin vivo molecular imaging in freely-moving rats. An achievedlow power dissipation of 161 µW allows its implantationwithout causing major traumas. The sensor can be operatedwith long operation time in order to minimize overall datarate, which simplifies mechanically and electrically theexternal control system. Laboratory measurements showedthat with an integration time of 1 s no hit losses were observedwithin activity ranges expected for the final experiments.

Further laboratory measurements will be performed inaqueous solutions containing radio-tracers of biologicalinterests (18F, 11C, and 15O). The project will then proceedwith the integration of the sensor as a bio-compatible probeinto the final setup of the MAPSSIC project.

Another IMIC 10 mm x 1 mm probe size has beensubmitted, which is implementing a Depleted MonolithicActive Pixel Sensor (DMAPS)[16] approach. This chip isproduced on a 2 kΩ · cm wafer, and allows to get a pixelateddepleted sensor thanks to the triple-well option. Indeed, every100 µm x 50 µm pixel has a depleted sensor by applyinghigh-voltage to the substrate. The 100 µm thickness of theprobe with its 20 µm depleted sensor depth will help β-ray detection and γ-ray rejection. Analog monitoring of eachpixel preamplifier and digital readout of the probe will provideadvanced solutions to characterize and operate the needle. ThisDMAPS chip will be tested in 2017.

REFERENCES

[1] S.R. Cherry. (2011) Functional whole-brain imaging in behaving rodents.Nat. Methods. 8(4):301-303.

[2] D. Schulz, P. Vaska. (2011) Integrating PET with behavioral neuroscienceusing RatCAP tomography. Reviews in neurosciences. 22(6):647-655.

[3] V.D. Patel et al. (2008) Imaging dopamine release with Positron Emis-sion Tomography (PET) and 11C-raclopride in freely moving animals.NeuroImage. 41(3):1051-1066.

[4] D. Schulz et al. (2001) Simultaneous assessment of rodent behaviorand neurochemistry using a miniature positron emission tomograph. Nat.Methods. 8(4):347-352.

[5] A.Z. Khyme et al. (2008) Real-time 3D motion tracking for small animalbrain PET. Phys. Med. Biol. 53(10):2651-2666.

[6] M. G. Spangler-Bickell et al. (2016) Optimising rigid motion compensa-tion for small animal brain PET imaging. Phys. Med. Biol. 61:7074-7091.

[7] D. Mannweiler et al. (2016) Imaging of Freely-Moving Mice in PET UsingPartially Rigid Transformations During List-Mode Reconstruction. IEEEMIC 2016 M15-5.

[8] F. Pain et al. (2000) SIC, an intracerebral radiosensitive probe forin vivo neuropharmacology investigations in awake small laboratoryanimals: Theoretical considerations and physical characteristics. IEEETrans. Nucl. Sci. 47:25-32.

[9] B. Weber et al. (2003) Quantitative CBF measurements in the rat usinga beta-probe and H215O. J. Cereb. Blood Flow Metab. 23:1455-1460.

[10] C. Woody et al. (2002) A study of scintillation veta microprobes. IEEETrans. Nucl. Sci. 49(5):2208-2212.

[11] J. Maerk et al. (2013) A wireless beta-microprobe based on pixelatedsilicon for in vivo brain studies in freely moving rats. Phys. Med. Biol.58(13):4483-4500.

[12] J. Godart et al. (2010) PIXSIC: A pixellated Beta-Microprobe forKinetic Measurements of Radiotracers on Awake and Freely Moving SmallAnimals. IEEE Trans. Nucl. Sci. 57(3):998-1007.

[13] L. Balasse et al. (2015) PIXSIC, a pixelated β+-sensitive probe forradiopharmacological investigations in rat brain: binding studies with18F-MPPF. Mol. Imaging. 17(2):163-167.

[14] L. Balasse et al. (2015) PIXSIC: a wireless intracerebral radiosensitiveprobe in freely moving rats. Mol. Imaging. 14(43):484-489.

[15] M. Mager. (2016) ALPIDE, the Monolithic Active Pixel Sensor for theALICE ITS upgrade. Nucl. Instr. Meth. Phys. Res. A 824:434-436.

[16] T. Wang et al. (2015) Development of a Depleted Monolithic CMOSSensor in a 150 nm CMOS Technology for the ATLAS Inner TrackerUpgrade. arXiv:1611.01206.