A Millimeter-scale Single Charged Particle Dosimeter for Cancer Radiotherapy
Kyoungtae Lee, Jessica Scholey, Eric B. Norman, Inder K. Daftari, Kavita K. Mishra, Bruce A. Faddegon, Michel M. Maharbiz, Mekhail Anwar
11 c (cid:13) a r X i v : . [ phy s i c s . i n s - d e t ] M a y (cid:13) A Millimeter-scale Single Charged ParticleDosimeter for Cancer Radiotherapy
Kyoungtae Lee, Jessica Scholey, Eric B. Norman, Inder K. Daftari, Kavita K. Mishra, Bruce A. Faddegon, MichelM. Maharbiz, and Mekhail Anwar
Abstract —This paper presents a millimeter-scale CMOS 64 × × µ m diode measures energydeposition by a single charged particle in the depletion region,and the array design provides a large detection area of 512 × µ m . Instead of sensing the voltage drop caused by radiation, theproposed system measures the pulse width, i.e., the time it takesfor the voltage to return to its baseline. This obviates the needfor using power-hungry and large analog-to-digital converters.A prototype ASIC is fabricated in TSMC 65 nm LP CMOSprocess and consumes the average static power of 0.535 mWunder 1.2 V analog and digital power supply. The functionalityof the whole system is successfully verified in a clinical 67.5 MeVproton beam setting. To our’ knowledge, this is the first workto demonstrate single charged particle detection for implantable in-vivo dosimetry. I. I
NTRODUCTION M ORE than half of cancer patients are treated withionizing radiation, where the fundamental goal is todeposit sufficient energy (dose) to destroy the tumor cellsand stop their proliferation. A key challenge in radiotherapyis to target the tumor while imparting minimal damage tosurrounding normal tissues. Commonly-used external beamradiotherapy (EBRT) employs x-ray photons to deliver aradiation dose to the tumor. This well-established methodencounters several difficulties: 1) X-rays pass through thewhole body, leaving unwanted dose in healthy tissues. This canbe critical in pediatric cancer, for example, where secondarymalignancy results from peripheral dose; 2) The dose fromx-rays is highest near the surface, dropping a few percent percentimeter with depth.Due to these issues, charged particle radiotherapy hasadvantages over x-rays. Unlike photons, charged particles(such as protons and carbon ions) deposit the highest dosein a specific location at the end of their range (the Braggpeak), theoretically allowing dose to be delivered with higherprecision and with less peripheral dose than with x-rays (SeeFig. 1 (b)).Despite the advantages of charged particle therapy, a currentlimitation is knowing the exact location of the Bragg peak(range uncertainty) which is caused by a number of factors.First, patient movement such as respiratory motion shifts theBragg peak. Second, charged particle interactions occurringwithin the body depend heavily on tissue atomic properties,
M. M. Maharbiz and M. Anwar contribute equally to this work as seniorauthors.
Charged-ParticleBeam (a)(b)
Real-TimeDose DataSensors Important OrganTumor Adjust the Beam R e l a t i v e D o s e Depth in Tissue
Proton, PristineProton, SOBPX-rayBragg PeakMargin Vital OrganTarget Tumor~1cm
Fig. 1: Illustrations of (1) cancer treatment with in-vivo dosimetry and (b) depth-dose curve for X-ray, proton (pris-tine), and proton (SOBP).which are difficult to determine accurately [1]. Lastly, day-to-day anatomical variations may make predictions inaccurate.The current clinical practice is to mitigate the range un-certainty by using a patient-specific maps of estimated par-ticle stopping power derived from CT image to predict thelocation of the Bragg peak. This method is rooted in thestoichiometric method [2], which provides a parametrized fit ofCT Hounsfield Units to the stopping power ratio of material.However, this comes at a price in range uncertainty due topotential errors in converting the X-ray attenuation coefficientto proton stopping power as well as uncertainties in the patientCT image. In addition, stoichiometric calibration cannot solvedaily anatomical variations and patient movement issues. Asa result, the typical range uncertainty is about 2.5 % of thetotal range. For example, if the Bragg peak is predicted to fall
100 mm inside a patient body, a range uncertainty of . (cid:13) will significantly impact the precision of the dose delivery.Given this, it is common in clinics to widen the Bragg peak tocover the full target volume, and then add treatment margins toensure the target is covered with prescription dose, resultingin increased dose to normal tissue. In addition, sub-optimalbeam arrangements may be selected to avoid delivering doseto a critical organ just distal to where the proton beam stops.An example of a spread-out Bragg peak (SOBP) and theadditional margins added to account for this range uncertaintyis illustrated in Fig. 1 (b)).Real-time in-vivo dosimetry (IVD) ameliorates uncertaintyby measuring the dose delivered in the body, potentiallyleading to more effective and safer closed-loop treatments(Fig. 1 (a)). Clinically viable IVDs have several importantconstraints. They must be millimeter scale for implantationthrough a standard core-biopsy needle; consume very smallamounts of power; have single-particle sensitivity; be capableof real-time measurement of energy deposition; and be suitablefor bio-compatible chronic implantation (usually 1-8 weeks)with appropriate medical-grade packaging. These requirementsstrongly drive the need for a CMOS platform capable of com-pact integration of low-power sensors and readout circuitry.While existing approaches have made progress towardsminiaturized IVD, no previous work has satisfied all require-ments [3]–[8]. Single MOSFET dosimeters have been themost widely used, as they can be easily fabricated in a smallsize [3], [4]. Integrated damage by radiation in the SiO layer of the MOSFET decreases the threshold voltage linearly.However, the lifetime is finite and it lacks single particle sen-sitivity due to the cumulative nature of the radiation induceddamages. Plastic scintillator, thermo-luminescent, or radio-luminescent dosimeters detect light intensity when a radio-sensitive material is exposed to radiation [5], [8]. However,they measure only cumulative dose, cannot provide real-timedata, and require bulky optical equipment to measure light thatprecludes implantation. Floating gate dosimeters measure thecurrent change when charges are trapped in the floating gateby radiation [6], but they lack single particle sensitivity.Most importantly, conventional dosimeters measure averagedose, and ignore a critical phenomenon: for a given dose,a single high linear-energy-transfer (LET, energy depositionper unit length) particle has a significantly different biologicaleffect on tissue than that of several low LET particles [9],[10]. The key metric to the biological effect is the energydeposition by each particle. Because the biological effectversus energy deposition is non-linear, the cumulative damagedoes not represent the true biological effect by radiation. Forexample, the normalized average number of lethal lesions ina HF19 human diploid fibroblasts cell produced by a single , , and
70 keV /µ m LET alpha-particle is approximately , . , and . , respectively [9]. The biological effect increasesmore rapidly than the energy deposition. The biological effectplateaus after
100 keV /µ m . Due to this non-linearity rela-tionship between the energy deposition and the biologicaleffect, single particle detection will be a key feature for next-generation IVDs, enabling analysis of the true biological effectby radiation.In this work, we solve these challenges by introducing a 64 ×
64 millimeter scale single charged particle CMOSdosimeter, compatible with in-vivo implantation for EBRT. Tothe best of our’ knowledge, the proposed system is the firstwork to enable single charged particle detection using onlyconventional CMOS chip fabrication process.II. T
HEORY OF OPERATION
This section describes how protons interact and depositenergy in matter, and how the deposited energy relates to thebiological effect. The expected signal measured by a diode isanalyzed. Finally, the acquisition of a pulse width (as opposedto a voltage level measurement) is discussed.
A. Proton interaction with matter
With the clinically-relevant energy range, protons depositenergy when passing through matter by three types of in-teractions: 1) Coulomb interactions with atomic electrons; 2)Coulomb interactions with atomic nuclei; and 3) nuclear reac-tions accompanied by creation of secondary particles (proton,neutron, electron, and gamma ray) [1]. The first type is themost dominant type of interaction, where a proton ionizesmatter, transferring part of its energy to electrons that deposittheir energy in proximity to the point of ionization ( ∼ ).The second type alters the proton trajectory and contributesto proton scattering. The last type is the rarest. In the firsttype, LET describes the average amount of proton energydeposited per unit length, and is well-modeled by the Bethe-Bloch equation. dE dep dx ∝ ρ ZA β (cid:20) ln 2 m e c γ β I − β − δ − CZ (cid:21) , (1)where dE dep /dx is the energy deposition per unit length, ρ is the density of the absorbing material, Z is the atomicnumber of the absorbing material, A is the atomic weight ofthe absorbing material, β = v/c where v is the velocity of theproton and c is the speed of light, m e is the electron mass, γ = (1 − β ) − / , I is the average ionization potential of theabsorbing material, δ is the density correction term, and C isthe shell correction term. Eq. 1 shows why it is challengingto predict the location of the Bragg peak, as the LET valueheavily depends on the material property and proton energy.Dose ( Gy = J / kg ) is widely used in clinical applicationsto quantify the radiation effect on tissue: Dose = N (cid:88) i =1 E dep,i m = E [ E dep ] × Nm , (2)where N is the number of protons, E dep,i is the energydeposition by each proton in the material, and m is the massof material where the energy deposition occurred. Dose is thesum of individual energy depositions per unit mass. However,the actual biological effect (e.g., the number of double strandbreaks in the DNA or cell mortality rate) for particles withhigher LET has a highly non-linear relationship with the E dep,i [9], [11]. This means that dose alone is an insufficient measureto evaluate the true effect on tissue. We also need the LET;that is, the single particle detection sensitivity. (cid:13) dep P-type Depletion Region EHP S ContactsSingleCharged-Particle + + + C par V drop Fig. 2: Illustration of diode sensing mechanism. (a) (b) th d r o p V Fig. 3: (a) Expected voltage drop at the diode sensing nodeassuming . µ m depletion thickness and . C par . TheLET data are retrieved from the NIST Pstar table [24]. (b)PW sensing diagram. B. Proton detection using a diode
When a proton interacts with a semiconductor diode, someof the energy deposited in the depletion region of the diodegenerates electron-hole pairs (EHPs). The average number ofEHPs generated in a silicon diode is
EHP = LET × t dep × qu sin θ p × .
12 eV , (3)where LET is dE dep /dx , t dep is the thickness of the depletionregion, θ p is the incident angle, and qu is the quenching effectthat describes approximately / of the deposited energy isused to generate EHPs. The other / is either dissipated byheat or via fast recombination of EHPs. The value .
12 eV represents the bandgap energy of the silicon. LET is a highlynon-linear function of proton energy. Note that because theproton beam angle from the source is fixed and we know thesensor orientation, mean θ p can be easily identified. Upper bound for std(PW)
Fig. 4: Monte Carlo simulated mean PW versus standarddeviation and PNR.Fig. 2 depicts the diode sensing mechanism. When the diodeis reversely biased by a current source, the generated electronsmove to the parasitic capacitance and create a voltage drop of V drop = q e × EHPC par , (4)where q e is the charge of an electron and C par is the parasiticcapacitance. Therefore, to achieve single particle sensitivity, anearly minimum size diode ( µ m × µ m ) is used to reduce C par because, for a single particle traversing the diode, theaverage number of EHPs is determined mostly by fabricationparameters and proton energy. In order to have wide detectionarea, we designed diodes into arrays. When designing an array,we want to maximize the fill factor (defined as the ratio ofdiode area to the area of the whole circuitry) to capture asmany incident particles as possible. Fig. 3 (a) depicts theaverage voltage drop by a single proton assuming C par of . , t dep of . , and a quenching effect of / . TheNational Institute of Science and Technology (NIST) pstartable is used to calculate the LET [24]. The voltage signalproduced during a collision ranges from to
78 mV at ∼
67 MeV proton energy range. The voltage drop is a nearlyinstantaneous event.Sensing the instantaneous voltage drop generated duringa collision requires high speed analog-to-digital converterswhich are power-hungry and occupy a large area (especiallyfor an array). In contrast, measuring the time it takes forthe generated voltage to return to its baseline is relativelystraightforward. We call this delay the pulse width (PW) (SeeFig. 3 (b)). The PW can be expressed as
P W = τ ln (cid:18) V drop V th (cid:19) = τ ln (cid:18) q e × qu × E dep . × sin θ p × V th × C par (cid:19) , where τ is the time constant at the diode sensing node and V th is the threshold voltage of detection. Even though the sensoroutput has a logarithmic relationship with E dep , this can bepre-calibrated before use.Including electronic noise at the diode sensing node, v n ,PW can be expressed as P W = τ ln (cid:18) V drop V th + v n (cid:19) . (5) (cid:13) Di ff Amp
30 fF20 fF V DCHPF (O ff -chip)LS p INV1LS n INV2 INV3
17 fF 17 fF V op V on Diode PDiode N V op DIS p DIS n V on EN p SW p SW n EN n V op V on E N p D I S p E N n D I S p DL AL V bDL V bAL Pixel unit X32 V bdio V bdio VDD
In+ In-V out V bAMP VDD V bLS V in V outWordline Wordline E N ( B i t l i n e ) D I S ( B i t l i n e _ n ) (a) (b) (c)(d) Fig. 5: Schematic diagram of (a) pixel unit, (b) differential amplifier, (c) level shifter (LS), and (d) in-pixel 1-bit SRAM.Given this, we can define pulse-width to noise ratio (PNR) asPNR = (cid:112) E [ P W ] σ P W > ln( V drop /V th ) σ v n /V th , (6)where the delta method is used to find the upper bound forthe standard deviation of a logarithmic function. Fig. 4 showsthe Monte Carlo simulated mean PW versus σ P W and PNR.Because PNR is proportional to the mean PW, V drop versusPNR is logarithmic. We can also define signal to noise ratio(SNR) as V drop /σ v n . Thus, the ratio of PNR to SNR isPNRSNR > V th ln( V drop /V th ) V drop , (7)which is always less than 1. This means that the PW sens-ing methodology loses resolution because of the logarithmictransformation of the signal. However, SNR in this analysisassumes perfect sampling of the critical time points of V drop (e.g. the time points corresponding to proton hits) which isimpossible in practical situations when using an analog-to-digital converter. The actual PNR loss is subsequently expectedto be less than that for the ideal situation.III. S YSTEM D ESIGN
The design consists of a 64 ×
64 pixel array, a main digitalblock, a SRAM control block, and a frequency locked loop(FLL). The system must feature low power consumptionfor future wireless applications, millimeter-scale size, enoughdetection area with sufficient fill-factor, and robustness toprocess mismatches.The following subsections discuss the analog pixel design,digital system design, FLL, and calibration steps.
A. Analog Pixel Design
Fig. 5 illustrates the pixel design. To reject common modenoise, a differential sensing scheme is used. Two diodes, diodeP and N, are grouped into one pixel unit. A P-type PN diode is used for maximizing the depletion region thickness. A nearlyminimum size PMOS current source supplies current to thediode. Changing the bias voltage of the current source ( V bdio )controls the depletion region thickness, time constant of thesensing node, and DC voltage.The differential amplifier should feature low input capac-itance, high gain, low noise, and low DC output voltagemismatch. Input transistors are critical, as there is a trade-off between the input capacitance and the DC output voltagemismatch. To balance these trade-offs, low V th (LVT) NMOSdevices with
600 nm /
600 nm are used as the input transistors.The differential amplifier occupies a µ m × . µ m area.A high-pass filter with a
30 fF
MOMCAP is used to rejectthe DC output voltage variance of the differential amplifier,and to set the DC voltage to a common voltage uniformlyacross all pixels by off-chip V DCHP F . To provide low f dB ,9 serial pseudo-resistors are used, as the f dB accuracy is notcritical.The level shifters ( LS ) shift the DC voltage downward( LS p ) or upward ( LS n ) to clip signals coming from the otherdiode. This enables passing signals from the correspondingdiode only. The V bLS is an essential variable that controls thetrade-off between sensitivity of signal detection ( V th ) and thepixel failure rate (i.e., the ratio between the number of failedpixels and the total number of pixels). For instance, lowering V bLSp increases the output DC voltage of LS p , leading tothe triggering of the following inverter by a smaller signal.However, it also increases the chance that the noise can triggerthe inverter.Diode transient pulse is converted to a digital pulse throughinverters. The output digital pulses, V op and V on , turn onPMOS switches to create the inverted signal on Data Line(DL), which is shared by pixels on the same row.Due to process, voltage, and temperature (PVT) variations,there is a chance that some pixels are constitutively active andoutput a false-positive signal even in the absence of radiationevents. Because the DL is shared by pixels on the same row, (cid:13) V op V on SW p SW n DLAL
Counter
TimeProton hits!
Pixel p hit! Pixel n hit! Violation 1 Violation 2
Fig. 6: Timing diagram.these false positives would hold the DL high and block signalscoming from other pixels. Therefore, an in-pixel standard 1-bit6T SRAM block is implemented to disable any false-positivepixel. Disabling these problematic pixels is called calibrationand will be explained in Section III-D.The overall pixel size is µ m × µ m , leading to a fill-factor of / . This means that there exists a high chanceof protons striking the transistors. Assuming the PN junctionsof the transistors have a similar depletion depth to that ofthe diode, this will create a voltage drop at the node withamplitude V drop ≤
250 mV × C par,diode C par,circuit , (8)where C par,diode and C par,circuit are the parasitic capacitancesat the diode node and the node of the proton hit, respectively.To address this issue, we designed the pixel such that either:1) the time constant of the node is less than the LSB ofthe PW sampling, which is µ s ; or 2) C par,circuit is muchgreater than C par,diode . For instance, the static DC current ofthe differential amplifier is
120 nA , and thus the worst casescenario will create a signal
P W = V drop dV /dt = 250 mV × C diode
60 nA ≈
10 ns . (9)Also, the parasitic capacitance at the HPF is more than 10times larger than that at the diode node so that the proton hitat the HPF cannot trigger the inverter.Fig. 6 illustrates the timing diagram. A proton hit at thediode creates a voltage pulse at one of the output nodes: V op or V on . Then, the same digital pulse but with opposite polarityis created on the DL. The PW is then quantized by a 10-bit digital counter with LSB of µ s . Sweeping signals ( SW p and SW n ) are 64 non-overlapping periodic
500 ns signalsgenerated from the main digital block. The sweeping signalsare then transferred to the Address Lines (ALs), which are alsoshared by the pixels on the same row, when the correspondingpixel has a proton hit and the sweeping pulse exists. Therefore,the main digital block can identify the column address of theproton hit by comparing the sweeping pulse and the AL signal.The sweeping lines are designed so that the overall delay ittakes for the signal to travel from the main digital to the pixeland back does not exceed
500 ns . ㅋ ㅋ DL[0]AL[0]
Pixel (0,0)
Pixel (0,1)
Pixel (0,2)
Pixel (0,63) ㅋ ㅋ
DL[1]AL[1]
Pixel (1,0)
Pixel (1,1)
Pixel (1,2)
Pixel (1,63) ㅋ ㅋ
DL[63]AL[63]
Pixel (63,0)
Pixel (63,1)
Pixel (63,2)
Pixel (63,63)
PD PU counter
SW[0:63]ADDR sweepSW[0] SW[1] SW[2] SW[63]
Digital
SRAM controlFPGA & biasing
10 kHz beacon D L o u t W/R P r i o r i t y E n c o d e r F I F O P I S O SIPO D a t a o u t Addressing counter counter
FLL
ASIC
Fig. 7: Overall system diagram.Note that this design methodology cannot distinguish mul-tiple proton hits at different pixels on the same row. Thisevent creates the DL pulse that is the logical OR operation oftwo voltage output pulses, making the DL pulse inseparable.However, such an event can be easily identified because morethan two sweeping pulses will be transferred onto the ALduring a single DL signal (See Violation 1 in Fig. 6). We cansimply discard these events because: 1) this is a rare event;and 2) discarding them will not change the overall statisticsbecause this is a purely stochastic event. Also, the DL PWmust be greater than
500 ns ×
64 = 32 µ s to guarantee thatthe column address is accurately identified. If not, there existsa chance that the column address is missing (See Violation2 in Fig. 6). Two types of events, the multiple hit eventand the address missing event, are called violation events,and we discard them. The multiple hit violations can bereduced by decreasing the time constant of the diode sensingnode. However, this increases the overall power consump-tion because the 10-bit counter must count faster. We canalso mitigate the address missing violations by sweeping thecolumns faster. Nevertheless, this might result in addressingthe wrong columns, as the overall delay of the sweeping signalcan exceed its pulse width. Note that the proton hit count canbe retrieved even when violation events occur.A key advantage of our method is that the pixels consumeonly static power in the absence of radiation. Unlike traditionalimaging applications where every pixel captures the signalperiodically, only the struck pixel captures the signal andconsumes dynamic power. This is made possible by the PWsensing strategy, and would not be true of a voltage sensingscheme. B. Digital System Design
Fig. 7 illustrates the overall system. The main digital blockfeatures the acquisition of the DL and AL signals, collecting (cid:13)
TABLE I: Internal Clock Configurations
Configuration Default Value Min Max t mst . , . , . , . , µ s 0 . µ s 0 . µ s 1 µ s t cnt t mst × , , , ,
128 6 µ s 1 . µ s 128 µ s t addr t mst × , , ,
32 0 . µ s 0 . µ s 32 µ s t PISO t mst × , , , . µ s 0 . µ s 8 µ s t SRAM t mst × , , ,
64 1 µ s 0 . µ s 64 µ s them and buffering, and configuring internal parameters. TheSRAM control block manages the enabling and disabling ofpixels, as well as the reading of current SRAM values.The DL and AL lines on each row have pull-down and pull-up transistors, respectively. The 10-bit counter starts countingat the rising edge of the DL signal. During counting, theaddressing block stores the current SW idx value when the ALis high. The counting is finished at the falling edge; and the10-bit counter value, 6-bit row address, 6-bit column address,and 2-bit status (00 : valid, 01: multiple proton hit, 10 : columnaddress missing) are transferred to the first-in-first-out (FIFO)block.When multiple rows have data, the priority encoder selectsa row that has data and the highest priority. The rows rangefrom 0 to 63, and a lower number translates to higher priority.This prevents data congestion at the interface between the 64row blocks and the FIFO. The FIFO block has a width of 24bits and depth of 16. Finally, a parallel-in serial-out (PISO)block receives the data from the FIFO and outputs each datato off-chip.The maximum latency happens when all rows have dataready and the FIFO is full. Therefore, the best and worst caselatency can be expressed as t latency,min ≈ P W + t cnt + t F IF O + 48 t P ISO t latency,max ≈ P W + t cnt + t F IF O + 48(16 + 64) t P ISO , where PW is the pulse width of the data and t cnt , t F IF O ,and t P ISO are the clock periods of the counter, FIFO, andPISO, respectively. In default settings, t latency ranges from P W + 36 µ s to P W + 1932 µ s . Also, since the PISO block isthe bottleneck in transferring data, the maximum proton fluxthat the digital block can handle is about 41,000 particles persecond.To give more flexibility of operation, internal parameterscan be configured through the serial-in parallel-out (SIPO)from off-chip. t mst , t addr , and t SRAM , which are the periodsof the main clock, the addressing clock, and SRAM clock,respectively, as well as t cnt and t P ISO can be configured asshown in Table I.
C. FLL
All internal digital clocks are generated from an on-chipFLL. The FLL obviates the need for external bulky crystaloscillators [16]. A
10 kHz beacon signal is sent from off-chip, and the FLL counts the digitally-controlled oscillator(DCO) clock during each period of the beacon signal. DCOfrequency is adjusted through negative feedback based on thedifference between the desired number of clocks in one period pbLSp opop p bLSp nbLSn onon n bLSn
900 950 1000 1050 1100 V bLSp,n (mV) E N ( % ) V bLSp V bLSn (a)(b) Fig. 8: (a) Illustrations of f p ( V bLSp ) and f n ( V bLSn ) and(b) measured percentage of enabled pixels versus V bLSp and V bLSn . V bLSp,n means V bLSp or V bLSn .and the actual counter value. This enables the generation ofa ∼
10 MHz main digital clock with approximately
280 kHz frequency resolution.
D. Calibration
The aforementioned constitutively active pixels are disabledbefore radiation through calibration steps. Each DL can bemonitored off-chip through a 64 to 1 multiplexer. The cali-bration steps are: 1) disable every pixel through the in-pixelSRAM; 2) enable one pixel and monitor the corresponding DLsignal for
50 ms ; 3) disable the pixel if the DL signal is noisyor high; and 4) repeat this process for the remaining pixels.This calibration process is carried out by an external FPGA,and takes approximately 5 minutes.By using this technique, we can indirectly measure thestatistics of mismatch among pixels. Fig. 8(a) depicts thefunction f p ( V bLSp ) and f n ( V bLSn ) . The percentage of enabledpixels after the calibration, EN , will be varied based on V bLSp,n value. EN is essentially the percentage of pixelswhose V op,n is VDD, which can be described asEN := 100 N (cid:88) i =1 { V op,n = V DD } /N = E [1 { V op,n = V DD } ] , where N is the total number of pixels and V op,n means V op or V on . Fig. 8(b) shows the measured EN after the calibrationwhen V bLSp and V bLSn are swept from
900 mV to .The slope from
975 mV to represents the mismatchof the function V op,n = f p,n ( V bLSp,n ) among the pixels. Forinstance, if the pixels were identical without any mismatch,the slope would be infinite because every pixel becomesenabled at a certain V bLSp,n value; that is, the graph showsthe measured cumulative distribution function of f p,n . First-order Gaussian fitting of the derivatives of the graphs gives the (cid:13) means of and
995 mV and the standard deviationsof .
33 mV and .
85 mV for f p ( V bLSp ) and f n ( V bLSn ) ,respectively. IV. M EASUREMENT R ESULTS
A prototype single charged particle dosimeter system wasfabricated in TSMC
65 nm
Low-power CMOS technology.The ASIC is µ m × µ m and its die photo is shown inFig. 9. The detection area is µ m × µ m with fill factorof / .This section describes the measurement setups and results.To analyze the electrical noise and pixel-to-pixel variations, aseparate 16 ×
16 testing chip was measured. The whole systemwas verified under a . proton beam generated by a76-inch cyclotron. The measurement results were compared tothose of the Monte Carlo simulation results. A. Electrical Measurement Results
Fig. 10 depicts the electrical measurement setup. A 16 ×
200 nm /
60 nm . The function generator generates 5,000 iden-tical pulses with µ s pulse width. These pulses cause aninstantaneous voltage drop at the diode node; the rest of thepixel circuitry outputs digital pulses on the DL. The meanand standard deviation of the PW were then measured. Thesemeasurements were repeated for various amplitude values ofthe input pulses.Fig. 11(a) shows the single pixel measurement results. ThePW variation increases with the mean PW, and the standarddeviation of the PW signals is less than µ s throughout thewhole operating range (0- ). PNR is 27.7 dB at PW.To measure pixel-to-pixel variation, 5,000 identical pulseswere presented to pixels and mean PWs were measured. Fig.Fig. 9: Chip die photo.
FunctionGenerator5000 PulsesAmp V bdio Fig. 10: Electrical measurement setup diagram. (a) (b)
Fig. 11: Electrical measurement result of (a) single pixel noiseand (b) pixel-to-pixel variation.
WaterPiston
Thickness adjustable water chamber d Ionization chamber 2 Patient shield & collimator
PCB & ASICFPGA PC
Patient treatment room
Fig. 12: Proton measurement setup diagram at CNL.10(b) is the normalized histogram of mean PWs of the 256pixels. This histogram is essentially the input-to-output PWgain mismatch among the pixels. The variation is mainly dueto the mismatch in the differential amplifier and the LS. Notethat because the input transistor (M1) mismatch, which isexpected to be significant due to small size, is embedded in themeasurement result, the actual pixel-to-pixel variation wouldbe smaller. In applications where high-spatial resolution dosemap is required, this variation can be pre-calibrated beforethe treatment by measuring each pixel’s responses at differentproton energies.
B. Proton Simulation and Measurement Results
The prototype ASIC was tested at Crocker Nuclear Labo-ratory (CNL) at University of California, Davis. The protonbeam facility has treated more than 1,700 ocular patients withmalignant and benign ocular tumors since 1994 [13]–[15]. (cid:13)
Raw beam current (nA) M ea s u r e d f l ux ( / s ec / mm ) V i o l a ti on r a t e ( % ) Clinical range (3-18 nA)
Fig. 13: Proton beam current versus measured flux and viola-tion rate.Fig. 12 depicts a simplified diagram of the proton mea-surement setup in an eye-treatment room. A . with . full width half maximum (FWHM) proton beamis generated by the 76-inch cyclotron. The beam enters thetreatment room and passes through: ionization chamber 1which monitors the dose; a thickness adjustable water chamberthat attenuates the proton beam energy; ionization chamber 2;a patient shield; and lastly a collimator. The ASIC was placedat the position of the patient’s eye during treatment (the iso-center) and the data was collected via an FPGA. The beamenergy at the patient was controlled by the water chamber. Dueto the nature of the energy loss mechanism of the chargedparticle (Eq. (1)), the energy deposition by protons has aninverse relationship with the proton energy above ∼ . ,thus EHP increases with increasing water thickness. Also, asthe water thickness increases, the beam scatters more and thusfewer protons reach the detector. This leads to a smaller protonflux, which is the number of protons in unit area per second,at the detector. These relationships are summarized in TableII.Fig. 13 shows measured proton flux and violation rate atdifferent proton beam current settings. The raw beam currentis proportional to the actual proton flux. The measured fluxincreases linearly from . to , and starts to saturateafter . Even though the clinical range at CNL is from to
18 nA , the remaining inviolate data provides enoughdata to extract meaningful statistics of energy deposition. Wecan also decrease the time constant at the diode sensing nodeTABLE II:
Relationship between the water thickness (d in Fig. 12)and the beam characteristics.
Water Thickness (d) Increase Decrease
Proton energy loss in the water chamber Increase DecreaseProton energy after the water chamber Decrease IncreaseProton energy deposition in the detector Increase DecreaseProton flux at the detector Decrease Increase d = 0 mm d = 10 mm(a) (b) (c)d = 20 mmd = 25 mm d = 27 mm d = 29 mm(d) (e) (f)
Count = 83,950Mean = 721 usStd = 502 us Count = 54,904Mean = 810 usStd = 540 us Count = 37,343Mean = 974 usStd = 646 usCount = 29,333Mean = 1,159 usStd = 828 us Count = 26,993Mean = 1,271 usStd = 1,069 us Count = 6,222Mean = 1,421 usStd = 1,432 us
Fig. 14: Normalized PW histograms measured for 80 secondsat (a) d = , (b) d =
10 mm , (c) d =
20 mm , (d) d =
25 mm , (e) d =
27 mm , and (f) d =
29 mm .to reduce the chance of having violations.Normalized PW histograms measured for 80 seconds of thebeam time at different water thicknesses are shown in Fig.14. The 10-bit counter quantized the PW of the DL signalsfrom 0- µ s with µ s resolution. Any DL signal whosePW is more than µ s is considered to be saturated. Thetotal proton count decreases as the water thickness increases,mainly due to the proton scattering in the water chamber. Asexpected, the mean PW, which indirectly measures the meanenergy deposition in the depletion region, increases as thewater becomes thicker.The histograms are rightward-skewed and become wider asthe water thickness increases. This is mainly due to the Landaueffect [17], which is the fluctuation in energy loss by ionizationof fast charged particles in a thin layer of matter. This isessentially what the sensor measures: energy loss (PW) byionization (generation of EHPs) of a charged particle (proton)in a thin layer of matter (depletion region).To verify the proton measurement data, the Tool for ParticleSimulation (TOPAS) was used [18]–[20]. TOPAS wraps andextends the Geometry and Tracking 4 (GEANT4) Monte Carloparticle simulator. GEANT4 is an industry gold-standard foranalyzing the behavior of atomic particles [21]–[23]. Fig. 15(a) shows the measured total number of protons, mean PW,and summation of PW over a 80-second window. The TOPAS (cid:13) (a) (b) Proton measurement result TOPAS simulation result
Fig. 15: (a) Measured total number of protons, average PW, and summation of PW for 80 seconds of beam time (b) TOPASsimulated total number of protons, average energy deposition, and total energy deposition.TABLE III: Comparison table with related state-of-the-art works. [3] [4] [5] [6]
This work
Sensing method Vth shift Vth shift RL/OSL Floating gate DiodeSensing area (mm or mm ) 0.3 × × × × × × Single particle detection? No No No No Yes simulated total number of protons, mean energy deposition(average energy deposited by each proton), and total energydeposition for all protons, assuming a depletion region thick-ness of . µ m , is shown in Fig. 15 (b). The trends matchwell with each other, and we can plot the measured mean PW(orange graph in Fig. 15 (a)) in the x -axis and the TOPASsimulated mean energy deposition (orange graph in Fig. 15(b)) in the y -axis to show the energy deposition versus PWrelationship (See Fig. 16). As expected from Eq. 5, thesehave a logarithmic relationship. The time constant at the diodesensing node is estimated to be µ s from a curve fit of thedata.The prototype ASIC consumes average static powers of µ W , µ W , and µ W for the analog pixel array, digitalsystem, and FLL, respectively. A comparison table with state-of-the-art dosimeters are summarized in Table III. This workhas the second largest sensing area of . × .
512 mm ,the lowest power consumption among active sensors, and thecapability of detecting radiation in real time. Most importantly,this work is the first work that can detect energy depositionby single charged particles with a form-factor and the powerconsumption compatible with wireless in-vivo dosimeter forcancer therapy. This work enables not only the detection ofthe Bragg peak, but an analysis of the radiation dose’s truebiological effect. Average energy deposited in TOPAS (KeV) A v e r a g e P W i n m ea s u r e m e n t ( u s ) MeasurementLogarithmic curve fitting
Fig. 16: Topas simulated average energy deposition versusmeasured average PW.V. C
ONCLUSION
A new CMOS diode based 64 ×
64 single charged particleradiation detector is proposed and verified using a clinicalproton beam. The design incorporates an analog pixel arraywith nearly minimum sized diodes, a digital system, SRAMcontrol block, and FLL. Theoretical analysis of measuringcharged particles using a diode is presented. The prototype isabout × with a detection area of . × .
512 mm .The proton measurement results are compared with detailedsimulation results. We envision that the proposed system (cid:13) can be used for various cancer therapies, including targetedradionuclide therapy or hadron beam therapy.A CKNOWLEDGMENT
The authors would like to thank the Chan ZuckerburgBiohub, the Department of Radiation Oncology at UCSF,Crocker Nuclear Laboratory (CNL), Berkeley Sensors andActuators Center (BSAC), Berkeley Wireless Research Center(BWRC), Sang Min Han of University of California, Berkeleyfor helpful discussions, and Hyun Joo Song of University ofCalifornia, Berkeley for her help in the preparation of thismanuscript. R
EFERENCES[1] W. Newhauser et al., “The Physics of Proton Therapy,”
Physics inMedicine and Biology , vol. 60, pp. R155-R209, 2015.[2] M. Yang et al., “Comprehensive analysis of proton range uncertaintiesrelated to patient stopping-power-ratio estimation using the stoichiometriccalibration,”
Physics in Medicine and Biology , vol. 57, pp. 4095-4115,2012.[3] G. Beyer et al., “An Implantable MOSFET Dosimeter for the Measure-ment of Radiation Dose in Tissue During Cancer Therapy,”
IEEE SensorsJournal , vol. 8, pp. 38-51, 2008.[4] E. Gurp et al., “In Vivo Dosimetry with a Linear MOSFET Array toEvaluate the Urethra Dose During Permanent Implant BrachytherapyUsing Iodine-125,”
International Journal of Radiation Oncology BiologyPhysics , vol. 75, no. 4, pp.1266-1272, 2009.[5] C. Andersen et al. “Characterization of a Fiber-coupled Al O :C Lumi-nescence Dosimetry System for Online In Vivo Dose Verification during Ir Brachytherapy,”
Medical Physics , vol. 36, pp. 708-718, 2009.[6] A. Shamim et al., “Wireless Dosimeter: System-on-Chip Versus System-in-Package for Biomedical and Space Applications,”
IEEE Transactionson Circuits and Systems II , vol. 55, pp. 643-647, 2008.[7] E. Grusell et al., “General Characteristics of the Use of Silicon DiodeDetectors for Clinical Dosimetry in Proton Beams,”
Physics in Medicineand Biology , vol. 45, pp. 2573-2582, 2000.[8] L. Wang et al., “Determination of the Quenching Correction Factors forPlastic Scintillation Detectors in Therapeutic High-energy Proton Beams,”
Physics in Medicine and Biology , vol. 57, pp.7767-7781, 2012.[9] D.T. Goodhead et al., “Mutation and Inactivation of Cultured MammalianCells Exposed to Beams of Accelerated Heavy Ions IV. BiophysicalInterpretation,”
International Journal of Radiation Biology , vol. 37(2),pp. 135-167, 1980.[10] D.T. Goodhead, “Initial Events in the Cellular Effects of Ionizing Ra-diations: Clustered Damage in DNA,”
International Journal of RadiationBiology , vol. 65, pp. 7-17, 1994.[11] E.J. Hall, “Radiobiology for the Radiologist, 7th Edition”
Hagerstown,Md. :Medical Dept., Haper & Row , 2012.[12] K. Lee et al., “A 64 ×
64 Implantable Real-Time Single-Charged-ParticleRadiation Detector for Cancer Therapy,”
IEEE International Solid-StateCircuits Conference (ISSCC) , 2020.[13] I.K. Daftari et al., “An Overview of the Control System for DoseDelivery at the UCSF Dedicated Ocular Proton Beam,”
InternationalJournal of Medical Physics , vol. 5, pp. 242-262, 2016.[14] I.K. Daftari et al., “Scintillator-CCD Camera System Light Output Re-sponse to Dosimetry Parameters for Proton Beam Range Measurement,”
Nuclear Instruments and Methods in Physics Research A , vol. 686, pp.7-14, 2012.[15] B.A. Faddegon et al., “Experimental Depth Dose Curves of a 67.5MeV Proton Beam for Benchmarking and Validation of Monte CarloSimulation,”
Medical Physics , vol. 42, pp. 4199-4210, 2015.[16] W. Biederman et al., “A Fully-Integrated, Miniaturized ( .
125 mm ) . Wireless Neural Sensor,”
IEEE Journal of Solid-State Circuits ,vol. 48, pp. 960-970, 2013.[17] L. Landau, “On the Energy Loss of Fast Particles by Ionization,”
Journalof Physics , vol. 8, 1944.[18] TOPAS : tool for particle simulation : http://topasmc.org[19] J. Perl et al., “TOPAS - An Innovative Proton Monte Carlo Platformfor Research and Clinical Applications,”
Medical Physics , vol. 39, pp.6818-6837,2012. [20] B. Faddegon el al., “The TOPAS Tool for Particle Simulation, a MonteCarlo Simulation Tool for Physics, Biology and Clinical Research,”
Physica Medica: European Journal of Medical Physics , vol. 72, pp. 114-121, 2019.[21] S. Agostinelli et al., “GEANT4 - A Simulation Toolkit,”
NuclearInstruments and Methods in Physics Research Section A , vol. 506, pp.250-303, 2003.[22] J. Allison et al., “GEANT4 Developments and Applications,”
IEEETransactions on Nuclear Science , vol. 53, pp. 270-278, 2006.[23] J. Allison et al., “Recent Developments in GEANT4,”