Detector Electronics

DoE Basic Energy Sciences (BES) Neutron & Photon Detector Workshop August 1-3, 2012 Gaithersburg, Maryland Detector Electronics spieler@lbl.gov Detector System Tutorials at http://www-physics.lbl.gov/~spieler or simply web-search spieler detectors More detailed discussions in H. Spieler: Semiconductor Detector Systems, Oxford University Press, 2005

2 Many new detector systems require combinations of extreme High event rates Fast Timing High position resolution High energy resolution Single particle detection Although individual characteristics have been implemented, the combination of multiple performance levels often has not been demonstrated. Novel detectors often build on a range of different concepts. They are often viewed as impractical only demonstrated readout systems will be accepted

3 Detector Readout Functions Interaction Rates Position Sensing Energy Distribution Timing Some experiments require only one function, but many require a combination. Optimizing i.e. achieving the optimum electrical performance of all components is often not essential. Finding a compromise that achieves the required experiment performance is usually the goal. This does require understanding the experimental goals and cross-coupled contributions often the optimum solution is not the original concept.

4 Some Aspects of Function Requirements 1. Interaction Rates Signal-to-noise ratio If the signal-to-noise ratio is too low, counts of noise pulses will be significant Rate capability Count rates are limited by 2. Position Sensing Sensor collection times Electronic bandwidth Data readout rate Sensor segmentation Segment signal distribution Timing interactions at various distances, also within the sensor

5 3. Energy Distribution Energy resolution Sensor statistical fluctuations Electronic Noise Digitization accuracy 4. Timing Sensor charge collection time distribution measuring the current pulse may be better than the charge Electronic rise time i.e. bandwidth Changes in detector signal amplitude time walk Transport time fluctuations and jitter Minimum timing requires very different circuitry than energy measurements, but compromises with energy resolution are often practical. Timing variations due to changes in detector signal pulse shape. Timing shifts can also arise in the detector...

6 Varying delays in signal amplitude due to x-ray interaction location z 0 : Induced signal current depends on the coupling of the charge by its electric field to the electrode Initially, charge is induced over many strips. As the charge approaches the strips, the signal distributes over fewer strips. When the charge is close to the strips, the signal is concentrated over few strips With small electrodes the magnitude of the induced current increases greatly when the charge is close to the electrode. For x-rays this shifts the pulse arrival Note: Commonly presented energy conservation theories for induced charge are generally wrong. CURRENT (pa) 60 50 40 30 20 10 0 z 0 = 50 m z 0 = 200 m 0 1 2 3 4 TIME (ns)

7 Detector Systems Conflicts and Compromises Fast event rate High-speed electronics Increased electronic noise Degraded signal-to-noise ratio Alternative: Segmentation strips or pixels Reduce the event rate per channel Allows longer shaping time Reduces electronic noise Data readout rate can limit the event rate Record time intervals and on the readout chip assign to events Readout does not have to synchronize with incident hits

8 Position Sensing Segmentation strips or pixels Strips yield one-dimensional position sensing Position resolution is determined by strip-pitch and transverse diffusion.

9 Position resolution requires a low input impedance of the preamp: Amplifiers must have a low input impedance to reduce transfer of charge through capacitance to neighboring readout channels. C C C C ss ss ss ss STRIP DETECTOR C C C C C b b b b b For electrode pitches that are smaller than the bulk thickness, the capacitance is dominated by the fringing capacitance to the neighboring strips C SS. Example: 1 2 pf/cm for strip pitches of 25 100 m on Si. The backplane capacitance C b is typically 20% of the strip-to-strip capacitance.

The input impedance of the amplifier must be small at the relevant frequencies determined by the pulse shaper. Two different shapers with the same 100 ns peaking time: 10 1 1 MAGNITUDE 0.8 0.6 0.4 CR-RC SHAPER 100 ns PEAKING TIME MAGNITUDE 0.8 0.6 0.4 CR-4RC SHAPER 100 ns PEAKING TIME 0.2 0.2 0 0 4 8 12 16 20 FREQUENCY (MHz) 0 0 4 8 12 16 20 FREQUENCY (MHz) The CR-RC shaper appears narrower, but reaches to higher frequencies. The frequency range scales inversely with shaping time. Charge-Sensitive preamplifiers are commonly viewed as yielding a large effective input capacitance. In reality, they commonly have an input resistance.

11 Pulse Response of a Basic Amplifier A voltage step vi ( t) at the input causes a current step io( t) at the output of the transistor. For the output voltage to change, the capacitance The output voltage changes with a time constant RLCO where R is the output load resistance. L C O at the output must first charge up., FREQUENCY DOMAIN TIME DOMAIN log A A v v0 g m R L g m i Co INPUT OUTPUT V 0 A = 1 v 1 RLCo 0 log V = V 1 exp( t / ) = RLCo 0 ( ) UPPER CUTOFF FREQUENCY 2 f u The time constant corresponds to the upper cutoff frequency : 1 f 2 u

Input Impedance of a Charge-Sensitive Amplifier Input impedance Zf Zf Z i ( A 1) A 1 A For no amplifier phase shift, a feedback capacitor will yield a large input capacitance. However, amplifier gain vs. frequency beyond the upper cutoff frequency 0 A i 1 Feedback impedance Z f i C i 1 1 Input Impedance Zi C f 0 0C i Imaginary component vanishes low frequencies ( f < f u ): capacitive input high frequencies ( f > f u ): resistive input f DETECTOR Very many charge-sensitive amplifiers operate in the 90 phase shift regime. Resistive input C d f Q i v i C A f v o 12

13 However... Note that the input impedance varies with frequency. Example: cutoff frequencies at 10 khz and 100 MHz, low frequency gain = 10 3 OPEN LOOP GAIN AND PHASE INPUT IMPEDANCE (C f = 1 pf) 1000 200 10 6 0 OPEN LOOP GAIN Av0 100 10 1 0.1 0.01 PHASE GAIN 160 120 80 40 PHASE (deg) INPUT IMPEDANCE ( ) 10 5 10 4 10 3 PHASE IMPEDANCE -20-40 -60-80 PHASE (deg) 0.001 0 10 3 10 4 10 5 10 6 10 7 10 8 10 9 FREQUENCY (Hz) 10 2 10 3 10 4 10 5 10 6 10 7 10 8 10 9 FREQUENCY (Hz) -100 The relevant frequency range is determined by the frequency passband of the pulse shaper. This is 5 15 MHz for a typical 20 ns shaper, so in this example the ohmic input is effective at much longer shaping times.

14 In the resistive regime the input impedance Z i 1, C 0 f where Cf is the feedback capacitance and 0 is the extrapolated unity gain frequency in the 90 phase shift regime. OPEN LOOP GAIN Av0 1000 100 10 1 0.1 0.01 0.001 PHASE GAIN 0 10 3 10 4 10 5 10 6 10 7 10 8 10 9 FREQUENCY (Hz) 200 160 120 80 40 0 PHASE (deg) Low-power amplifiers with a gain-bandwidth product much greater than in this example are quite practical, so smaller feedback capacitances are also possible. Time Response of a Charge-Sensitive Amplifier Input resistance and detector capacitance form RC time constant: i i D 1 i C R C 0 f C D

15 Crossed strips provide 2-dimensional position sensing x y Fake hits at high rates

16 Problem: Ambiguities with multiple simultaneous hits ( ghosting ) HIT GHOST n hits in acceptance field n x-coordinates n y-coordinates n 2 combinations Pixels of which 2 n n are ghosts

17 Energy Distribution Pixels are also advantageous in achieving low electronic noise. Equivalent Noise Charge: TS Shaping Time 1 Q i FT e F C F A C 2 2 2 2 2 n n i S n v d vf f d TS 2 in Spectral noise current density, e.g. in 2eIbias strip length C Detector capacitance strip length en d Amplifier spectral noise voltage density e 2 1 n gm Fi, Fv, F vf "Shape Factors" that are determined by pulse shaper In many applications the noise voltage contribution dominates: The main noise source is within the transistor, forming an output noise current. Transferred to the input, the amplifier spectral noise voltage density e n results from dividing the output noise current by the transistor gain, i.e. the transconductance g. m How does transconductance depend on the current (power) of the input transistor?

18 In analog circuitry the current draw is driven by the requirements of noise and speed. dic di D Both depend on transconductance gm (BJT) or gm (FET). dv dv FET transconductance is a non-linear function of current (W =100, L= 0.8 m): BE GS 10-2 TRANSCONDUCTANCE (ms) 8 6 4 2 TRANSCONDUCTANCE (S) 10-3 10-4 10-5 10-6 10-7 0 0 2 4 6 DRAIN CURRENT I D (ma) 10-8 10-9 10-8 10-7 10-6 10-5 10-4 10-3 10-2 DRAIN CURRENT I D (A) Power efficiency depends on transconductance per unit current g / I. m D

19 Measurements on 0.8 m CMOS process for different channel lengths L 25 20 g m /I D (V -1 ) 15 10 L= 25.2 m L= 0.8 m 5 0 10-4 10-3 10-2 10-1 10 0 10 1 10 2 10 3 I D /W (A/m) For a given device the x-values are proportional to device current. e.g. for W 100 m, I D / W 10 corresponds to a current of 1 ma. Traditional detector front-ends were designed to minimize noise, but accepting a 3 to 5-fold increase in noise reduces power by orders of magnitude!

20 Scaling of transistor size to optimize power Procedure: For a small device select the current density for a given gm / I D from plot on previous page. Then increase device width while scaling the device current proportionally (maintain current density). increase transconductance (reduce noise) Minimum noise when C FET C det Q n (e) (capacitive matching) 10 4 10 3 10 2 10 1 g m /I D = 24 10-7 10-6 10-5 10-4 10-3 10-2 10-1 10 0 10 1 I D (A) 23 Cdet = 10 pf, CFET = 1 ff/ m 20 16 12 8 4 For larger device widths the increase in capacitance overrides the reduction in noise. This yields minimum noise, but is not most power efficient! For gm / I D 24, minimum noise of 1400 e at 50 A, but for gm / I D 20 a noise level of 1000 e is obtained at 30 A. Given noise level can be achieved at low and high current. 2 1

21 Noise Cross-Coupling C C C C ss ss ss ss STRIP DETECTOR C C C C C b b b b b In strip and pixel detectors the noise at the input of an amplifier cross-couples to its neighbors.

Cross-Coupling Function in Strip and Pixel Detectors The center amplifier s output noise voltage v no causes a current noise in to flow through its feedback capacitance Cf and the interelectrode capacitances into the neighboring amplifiers, adding to the other amplifiers noise. C b The backplane capacitance attenuates the signal transferred through the strip-to-strip capacitance C. ss The additional noise introduced into the neighbor channels vno 1 vno 1 vno2 2 1 2 C / C b ss NEIGHBOR 1 MEASUREMENT NEIGHBOR 2 STRIP Z i1 v v v For a backplane capacitance Cb Css /10 the amplifier s noise with contributions from both neighbors increases by 16%. In pixel detectors additional paths must be included. This requires realistic data on pixel-pixel capacitances (often needs tests). i n1 C ss i C n1 f i n1 no1 no no2 C i b b i i n n2 C ss C i f n Z i2 i n2 i C n2 f 22

23 Strip Detector Model for Noise Simulations Noise coupled from neighbor channels. Analyze signal and noise in center channel. Includes: a) Noise contributions from neighbor channels b) Signal transfer to neighbor channels c) Noise from distributed strip resistance (+ potential effect of strip inductance) d) Full SPICE model of preamplifiers 1600 Measured Noise of a Module Test beam experiment, so realistic environment: p-strips on n-bulk, BJT input transistor Simulation Results: 1460 el (150 A) 1230 el (300 A) No digital cross-talk Noise [rms el] 1500 1400 1300 1200 Noise can be predicted with good accuracy. 50 100 150 200 250 300 Current in Input Transistor [ A]

24 Advantages of Segmentation 1. Segmentation reduces detector capacitance lower noise for given power 2. Segmentation reduces the hit rate per channel longer shaping time, reduce voltage noise 3. Segmentation reduces the leakage current per channel (smaller detector volume) reduced shot noise 4. Segmentation allows higher overall event rates 5. Overall power per unit area can remain fairly constant with the increase in the number of segments, since the power per segment can reduce with smaller segments. Segmentation is a key concept in large-scale detector systems.

25 CCD a traditional pixel detector Signal charge deposited in a pixel is read out by shifting it through the neighboring pixels until it reaches the end. It is then transferred to the output amplifier. Multiple excited pixels are transferred sequentially, so all individual pixel signals are read out. Pixels can be very small, but readout is slowed by the sequential succession. Multiple readout column groups with output amplifiers together with increased clock rates can speed up the output. High-rate devices are read out in the MHz regime, which greatly increases the power dissipation. However, at high readout rates the power is dominated by the power loss in the readout clock lines. This is even a problem at sub-mhz rates: SERIAL OUTPUT REGISTER PIXEL ARRAY OUTPUT AMPLIFIER The VXD3 CCD operating at a pixel transfer rate of 200 khz and a clock level of 10 V had peak currents of 1.3 A, leading to potential cross-talk.

26 Pixels directly coupled to front-ends offer high rates and low noise The most flexible is the hybrid pixel device: SENSOR CHIP The sensor electrodes are patterned as a checkerboard and a matching two-dimensional array of readout electronics is connected a two-dimensional array of contacts, for example solder bumps. READOUT CHIP BUMP BONDS READOUT CONTROL CIRCUITRY WIRE-BOND PADS FOR DATA OUTPUT, POWER, AND CONTROL SIGNALS via Hybrid pixels allow independent optimization of sensor and readout, e.g. allows non-si sensor. Drawback: Engineering complexity much greater than for common chips.

27 Example: ATLAS pixel detector about 10 8 channels The initiating complex pixel design, which has been working reliably. Each pixel cell includes Charge-sensitive-amplifier + shaper per pixel Threshold comparator per pixel Trim-DAC per pixel for fine adjustment of threshold Time-over-threshold analog digitization Test pulse circuitry per pixel (dual range) Buffer memory to accommodate trigger latency Circuitry to mask bad pixels

28 ATLAS Pixel Cell DETECTOR PAD CHARGE-SENSING PREAMPLIFIER COMPARATOR LEADING + TRAILING EDGE RAM COLUMN BUS GLOBAL DAC LEVELS SERIAL CONTROL BUS DUAL RANGE CALIBRATION V TH ToT 40 MHz CLOCK FROM CALIBRATION DAC ToT TRIM DAC THRESHOLD TRIM DAC GLOBAL INPUTS AND CONTROL LOGIC 0.25 m CMOS, Q n 170 e 40 W per cell; total power for 2880 pixels: 200 mw (incl. peripheral circuitry)

29 Readout Scheme PIXEL CELLS COLUMN BUFFERS CONTENT ADDRESSABLE MEMORY CONTENT ADDRESSABLE MEMORY DATA OUT TRIGGER Pixels continuously active, but don t send signals until struck (self-triggered). Time stamp for struck pixels stored immediately in Content Addressable Memory Data stored in pixel until Level 1 trigger received for stored time stamp.

30 ATLAS Pixel Detector Fabrication IC components Pixel size: 50 m x 400 m Pixel IC Pixel IC Size is historical: could be 50 m x 200 m Power per pixel: < 40 W Each chip: 18 columns x 160 pixels (2880 pixels) Module size: 16.4 x 60.4 mm 2 16 front-end chips per module 46080 pixels per module Fabricated in 0.25 m CMOS ~ 3.5 10 6 transistors per chip Functional to > 100 Mrad Module Readout IC Support and Test ICs Radiation resistant to higher fluences than strips because low noise provides large performance reserves.

31 ATLAS Pixel Module Sensor used as substrate to mount 16 readout ICs READOUT CONTROLLER FLEX HYBRID SENSOR SOLDER BUMPS PIXEL ICs Two-dimensional arrays of solder bump bonds connect ICs to sensor. SENSOR SOLDER BUMP SIGNAL READOUT IC 50 m 400 m

32 Threshold dispersion must be smaller than noise. Small feature sizes large threshold dispersion correct with trim DAC Threshold dispersion before and after trimming THRESHOLD (e) THRESHOLD (e) 7000 6000 5000 4000 3000 2000 1000 5500 5000 4500 4000 3500 3000 0 10000 20000 30000 40000 PIXEL NUMBER = ROW + (160 x COLUMN) + (2880 x CHIP) 0 10000 20000 30000 40000 PIXEL NUMBER = ROW + (160 x COLUMN) + (2880 x CHIP) 620 e 0 2000 4000 6000 THRESHOLD (e) 60 e 3500 4000 4500 THRESHOLD (e)

33 Noise Distribution NOISE (e) 500 400 300 200 100 0 10000 20000 30000 40000 PIXEL NUMBER = ROW + (160 x COLUMN) + (2880 x CHIP) Three groups visible: 1. nominal pixels 2. Extended pixels that bridge columns between ICs (spikes every 2880 pixels) 3. Ganged pixels to bridge rows between ICs

34 Monolithic Pixels in Standard IC Designs Example: Electron Microscopy (P. Denes et al.) Essentially all charge collected from thin region for good position resolution Electron Detection Detector uses the passive region between the circuit layers and the base. Pixels are small (so there can be more of them, but they are less intelligent than hybrid pixels) But... Radiation damage (electric field in the detection region is not well controlled) Diffusion (because collection region is not depleted)

35 Multi-Tier Electronics (aka SOI or 3D ) CMOS Circuitry 7 µm Isolation Oxide Sensor Layer MIT Lincoln Lab 3-Tier Design (FNAL) 3 transistor levels 11 metal layers Accommodate additional circuitry for given pixel size. Also extensive SOI development at KEK (R. Lipton, FNAL)

After introduction in high-energy physics (LHC), hybrid pixel devices with complex electronic readouts are now applied in a variety applications, e.g. high-rate x-ray detection and medical imaging. The pixel size is limited by the area required by each electronic readout cell. Pixel sizes of 30 100 μm are practical today, depending on the complexity of the circuitry required in each pixel. The readout IC requires more area than the pixel array to accommodate the readout control and driver circuitry and additional bond pads for the external connections. Multiple readout ICs are needed to cover more than several cm 2, so in mounting multiple readout ICs on a larger sensor requires that the ICs are designed with small edge areas. Implementing this structure monolithically would be a great simplification and some work has proceeded in this direction. Appropriate cooling can deal with power dissipation, but heating of the sensor should be limited. Electronics associated with each pixel can perform signal acquisition and pulse shaping and also record timing and provide local storage, so readout does not have to synchronize with hit rates. However, pixel readout designs are more complex than strip detector or conventional designs. Requires multiple simulations and independent cross-checks to verify capabilities. 36

37 Radiation Damage For x-rays and low-energy gammas the main cause of radiation damage is charge buildup at oxide-si interfaces. The trapped positive charge attracts electrons in the adjacent active Si region. GATE OXIDE Si E ox ELECTRON-HOLE RECOMBINATION ELECTRON INJECTION FROM Si ELECTRON-HOLE PAIRS FORMED BY RADIATION HOLE TRAPS TRAPS CLEARED BY TUNNELING FROM Si adapted from Boesch et al. IEEE Trans. Nucl. Sci (1986) 1191 In detectors this can lead to leakage between adjacent electrodes and in electronics it leads to MOSFET bias shifts ( digital circuit failure).

38 MOSFET Threshold Shift vs. Oxide Thickness 10 3 10 2 - V / DOSE (V/Mrad) FB 10 10 10 1 0-1 V d 2 10-2 10-3 1 10 100 OXIDE THICKNESS (nm) Saks et al., IEEE Trans Nucl. Sci. NS-31 (1984) 1249 Deep sub-micron MOSFETs have appropriate oxide thicknesses, so their radiation resistance is good. Cross-coupling in the detector may be critical.

Grounding A Common Cause of Noise A popular recipe is the star ground, i.e. connecting all ground lines to a common point. Example: Integrated Circuit V+ 39 DETECTOR LOAD The output current is typically orders of magnitude greater than the input current (due to amplifier gain, load impedance). Combining all ground returns in one bond pad creates a shared impedance (inductance of bond wire). The voltage drop due to the output current couples to the input.

40 Separating the ground connections by current return paths routes currents away from the common impedance and constrains the extent of the output loop, which tends to carry the highest current. V+ LOAD DETECTOR V-

Summary 41 Interplay of many interacting contributions must be understood. Requires understanding the physics of the experiment, detector, readout, rather then merely following recipes. Physics requirements must be translated to engineering parameters. Many details interact, even in conceptually simple designs. For example, analysis in time and frequency domain Single-channel recipes tend to be incomplete. Overall interactions must be considered. Simulations may provide the correct answers, but they should be crosschecked. Multiple equivalent simulations often give different results.

Challenges 42 No silver bullets! Systems design is crucial in advanced detectors. It is essential to understand key aspects and their interactions. Key front-end issues don t require detailed electronics knowledge of circuits, but understanding of basic underlying physics is essential. Broad physics education required. U.S. physics departments commonly do not recognize the scientific aspects of instrumentation R&D. Many developments are essentially made as technician efforts, so the simplistic perspective doesn t accept that novel developments require a scientific approach. Emphasis on theory and mathematical techniques neglects understanding of physics and how to apply it to undefined multidimensional problems. Novel detectors often build on a range of different concepts. General detector R&D can build an efficient base for multiple applications.