1. L. Ratti a,c, E. Pozzati a,c, C. Andreoli a,c, M. Manghisoni b,c, V. Re b,c, V. Speziali a,c, G. Traversi b,c a Università degli Studi di Pavia b Università degli Studi di Bergamo c INFN Pavia International Linear Collider Workshop Valencia, 6-10 November 2006 CMOS Monolithic Sensors with Sparsified Readout and Time Stamping Capabilities for Vertexing Applications at the ILC

2. 2 ILC Workshop 2006 – Valencia, 6-10 November 2006 MAPS for vertexing applications Tracking and vertexing systems in future high luminosity colliders (ILC, SLHC, Super B-Factory) will be operated at high rate with low material budget to optimize position and momentum resolution An ambitious goal is to design MAPS with similar readout functionalities as in hybrid pixels (sparsification, time stamping, e.g. FPIX) A readout architecture with data sparsification would be a new feature which could give some advantages with respect to existing MAPS implementations decouple modularity from readout speed reduce digital power dissipation flexibility in dealing with possible luminosity and background changes during the experiment lifespan Modern VLSI CMOS processes (130 nm and below) could be exploited to increase the functionality in the elementary cell Monolithic active pixel sensors (MAPS) may provide the required low mass and high granularity

3. 3 ILC Workshop 2006 – Valencia, 6-10 November 2006 In triple-well CMOS processes a deep N-well is used to isolate N- channel MOSFETs from substrate noise Deep N-well structure NMOS PMOS P-substrate Buried N-type layer P-well Standard N-well P-epitaxial layer Such features were exploited in the development of deep N-well (DNW) MAPS devices A DNW is used to collect the charge released in the epitaxial layer A readout channel for capacitive detectors is used for Q-V conversion  gain decoupled from electrode capacitance NMOS devices of the analog section are built in the deep N-well Using a large detector area, PMOS devices may be included in the front- end design  charge collection inefficiency depending on the ratio of the DNW area to the area of all the N-wells (deep and standard) Deep N-well MAPS concept

4. 4 ILC Workshop 2006 – Valencia, 6-10 November 2006 PMOS NMOSP-well High sensitivity charge preamplifier with continuous reset RC-CR shaper with programmable peaking time (0.5, 1 and 2 μs) A threshold discriminator is used to drive a NOR latch featuring an external reset A couple of test structures were fabricated in a 130 nm, triple well, epitaxial layer CMOS technology by STM PreamplifierShaperDiscriminatorLatch Gm A(s) b0 b1 + CFCF C1C1 C2C2 VtVt RST VFVF ~43 m The first prototypes proved the capability of the sensor to collect charge from the epitaxial layer (G. Rizzo et al., “A novel monolithic active pixel detector in 0.13 μm triple well CMOS technology with pixel level analog processing”, NIMA vol. 565, pp. 195-201, 2006) Pixel level processor (Apsel family)

5. 5 ILC Workshop 2006 – Valencia, 6-10 November 2006 in out I CFCF iFiF e IN e CS E E1E1 E2E2 VFVF Series noise from current source Series noise from input device Parallel noise from the feedback network ENC=40 e - rms @C D =320 fF (900 μm 2 sensor area) Two chips recently submitted to study charge spreading in the epi-layer and charge collection efficiency (130 nm CMOS process by STM) Threshold dispersion about 40 e - rms Power dissipation about 60 μW (although decreasing the input device current by a factor of three does not change significantly the noise performances) 900 μ m 2 sensor area Analog performances

6. 6 ILC Workshop 2006 – Valencia, 6-10 November 2006 ILC is expected to feature a beam structure with 2820 crossings per train (1 ms), with a duty-cycle of 0.5% Chance of a single cell being hit twice in a bunch train is ≤ 1% for 20 μm x 20 μm or smaller pixels  pipeline with a depth of one is sufficient to record ≥ 99% of events with no ambiguity  data can be readout all together during the intertrain interval Maximum hit occupancy is assumed to be 0.03 particles/crossing/mm 2 3 hits/particle yield a hit rate of about 250 hits/train/mm 2 Although the occupancy is quite small, a time stamp may prove useful to complement information from other detectors and reconstruct hit patterns Sparsification reduces the amount of data sent off the chip and digital power dissipation A DNW MAPS sensor for ILC is being designed based on a token passing readout scheme suggested by R. Yarema (R. Yarema, “Fermilab Initiatives in 3D Integrated Circuits and SOI Design for HEP”, ILC VTX Workshop at Ringberg, May 2006) DNW MAPS for vertexing at the ILC

7. 7 ILC Workshop 2006 – Valencia, 6-10 November 2006 PreamplifierDiscriminator 22T 14T iFiF CFCF VtVt Preamplifier response to an 800 e - pulse Smaller area than in the Apsel prototypes  smaller detector capacitance ENC=25 e - rms@C D =100 fF Power consumption: about 5 μW Threshold dispersion: about 30 e - rms Features power-down capabilities for power saving Cell analog front-end

8. 8 ILC Workshop 2006 – Valencia, 6-10 November 2006 199 ms 60  s Switch-on transient Preamplifier pulse response 1.2 0.0 0.2 0.4 0.6 V ON OFF The analog section in the elementary cell can be switched off during the intertrain interval in order to save power (analog power is supposed to be predominant over digital) Based on circuit simulations, power cycling with at least 1% duty-cycle seems feasible Considering the power available for the ILC vertex detector (20 W) and the vertex barrel area (about 170000 mm 2 ), the available power/cell is ≥4.7 μW for a pitch of 20 μm or larger, compatible with the power dissipation features of the analog front-end Preamplifier pulse response Power cycling

9. 9 ILC Workshop 2006 – Valencia, 6-10 November 2006 OE OEb t1t2t3t4t5 t5int4int3int2int1in CPb CP D Q Qb WE hit hitb tokin tokrst tokout getb_enQ Qb Lat_en R S token passing core Get X bus From the time stamp counter To the time stamp buffer Cell CK Get Y bus Master Reset time stamp register 4T 10T 13T 20T 76T From the discriminator Cell CK hit latch bus control FF Includes a 5 bit time stamp register and the data sparsification logic During the bunch train period, the hit latch is set in each pixel that is hit When the pixel is hit, the content of the time stamp register gets frozen Cell digital section

10. 10 ILC Workshop 2006 – Valencia, 6-10 November 2006 X=1X=16 Y=1 Y=2 Y=16 X=2 Time Stamp Buffer 1 Time Stamp Buffer 2 Time Stamp Buffer 16 Cell (1,1) Cell (1,2) Cell (1,16) Cell (2,1) Cell (2,2) Cell (2,16) Cell (16,1) Cell (16,2) Cell (16,16) 555 555444 4 4 4 4 4 5 MUX Last token out First token in X Y T TkinTkout 111 Tkin TkoutTkinTkout TkinTkoutTkin TkoutTkinTkoutTkinTkoutTkin TS 1 1 1 Serial data output gXb gYb TS Readout CK gXb gYb gXb gYb gXb gYb gXb gYb gXb gYb gXb gYb gXb gYb gXb gYb Cell CK gXb=get_X_bus gYb=get_Y_bus TS=Time_Stamp Tkin=token_in Tkout=Token_out Digital readout scheme

11. 11 ILC Workshop 2006 – Valencia, 6-10 November 2006 Detection 0 123 4 3031 Discriminator output Hit Token in Token out Output enable Discriminator output Hit Token in Token out Output enable XYTS CELL (2,6) XYTS CELL (5,8) Data output Serial Data Output Cell CK Readout CELL (2,6) CELL (5,8) CELL (2,6) CELL (5,8) (next hit cell) first event second event Token arrives at cell (2,6) Cell (2,6) data are read out Cell (5,8) data are read out Token arrives at cell (5,8) Timing diagram

12. 12 ILC Workshop 2006 – Valencia, 6-10 November 2006 Time stamp register Preamplifier Discriminator DNW sensor Sparsification logic Token passing core Hit-latch Bus control FF Nand gate 25 m ILC DNW MAPS demonstrator A 16x16 MAPS demonstrator chip is being designed in the 130 nm bulk CMOS technology by STM Increase in the number of elements just requires larger X- and Y-registers and serializer Further pitch reduction might be achieved by: replacing the digital time stamp register with an analog one using a further scaled process, namely a 90 nm CMOS technology by STM, supposed to have substrate features similar to those of the 130 nm process

13. 13 ILC Workshop 2006 – Valencia, 6-10 November 2006 A 90 nm CMOS process might provide some advantages in the design of DNW MAPS with respect to the 130 nm process about a factor of two reduction in silicon area occupation better power/channel thermal noise trade off better 1/f noise performances Normalized power spectral densities for NMOS devices belonging to the STM 130 and 90 nm processes smaller dispersion in device parameters Experimental characterization of test structures in 90 nm CMOS process will provide useful information as to the design of DNW MAPS be careful with gate currents DNW MAPS in 90 nm CMOS

14. 14 ILC Workshop 2006 – Valencia, 6-10 November 2006 A DNW MAPS demonstrator aimed at vertexing applications at the ILC is being designed in a 130 nm bulk CMOS technology The chip is capable of Further activities are under way to optimize the elementary cell size and geometry, including simplification of the pixel cell logic MAPS development in a 90 nm CMOS technology physical device simulations Conclusions The chip is capable of sparsified readout, time stamp generation and is compatible with power cycling operation Simulations show good noise and threshold dispersion performances at a power dissipation close to the ILC vertex specifications