1. Development of an ASIC for reading out CCDS at the vertex detector of the International Linear Collider Presenter: Peter Murray ASIC Design Group Science & Technology Facilities Council Rutherford Appleton Laboratory Harwell Science & Innovation Campus Didcot Oxfordshire OX11 0QX United Kingdom

2. ILC will collide e+ e- beams at energies up to 500GeV. Interactions recorded by 2 detectors, each with a vertex detector. Vertex detector resolution < 5 microns, with low power and minimum material. Thus CCDs have been selected. Need to keep occupancy below 1% means that CCD pixel columns must be read out in parallel at 50Mhz. Low occupancy also makes on-chip data sparsification desirable. Detector overview

3. Already fabricated CPR chips Bump bond pads Wire/Bump bond pads CPR1 CPR2 Voltage and charge amplifiers 125 channels each Analogue test I/O Digital test I/O 5-bit flash ADCs on 20 μm pitch Cluster finding logic (2  2 kernel) Sparse readout circuitry FIFO Designed for readout of Column Parallel CCDs under development for LCFI (Linear Collider Flavour Identification) CPR2 Size : 6 mm  9.5 mm 0.25 μm CMOS process (IBM)

4. Chip reads out digitized pixel data only if it exceeds specified threshold. Charge from a particle may be shared between several CCD pixels. Hence sparsification logic sums data from every 2x2 array of pixels and compares with threshold. Chip also reads out data from pixels surrounding the hit pixels which triggered readout. Thus all interesting data guaranteed to be read out. Because of limited space for the logic the algorithm must be very simple. CPR2A Sparsification algorithm

5. Sparsification algorithm 2 by 2 cluster above threshold : 6 by 4 output data 11111111 111111 111111 11111111 11111111 11111111 11111111 11111111 222 22 1111 1111 1111 1111 1111 11111111 11111111 1111 2 22 1111 1111 1 1 1 1 1111 1111 CPR2A readout format – case 1

6. Output data format Colour coding: Address (H) Address(L) Gap (00000) Time-stamp Data Simplest case: 6 words of 5-bit data

7. Sparsification algorithm 4 by 2 input cluster, Output data: 6+8 by 4 111111 111111 111122 111122 111122 111122 111111 111111 11 11 11 111111111111 111 1111 1 2211 112211 112211 112211 11 11 111 1111 1 11 11 111 1111 1 11 11 111 1111 1 11 11 111 1111 1 111111 CPR2A readout format – case 2

8. Sparsification algorithm 111111 111122 111122 111111 111111 11 11 11 11 111 11 1 11 11 22 11 1111 1111 11 11 111 1111 1 11 11 111 1111 1 11 11 111 1111 1 CPR2A readout format – case 3 111111 111111 111111 111111 111111 111111 111111 111122 111122 111111 22 11 1 11 1 1111 1111 1111 11 11 111 1111 1 11 11 111 1111 1 11 1 11 1 1111 111111 11 11 11 1122 22 111111 2 input clusters, separated by 9 time steps Output data: 6+8+8 by 4 Maximum separation that is stored as a single data block

9. Sparsification algorithm When vertical cluster separation 10 data is stored as 2 separate data blocks.

10. Address (H) Address(L) Gap (00000) Time-stamp Data Extended data: 12 words in 2 groups Time-stamp repeated, no break in header signal Gap between groups of data, to identify new time-stamp Output data format

11. CPR2A Verilog simulations based on physics data (100 channels, 100 time steps) Near-perfect readout over 100 time steps (44 hit pixels, occupancy 0.44%) One missing data point in the output (channel 92, time stamp 4)

12. CPR2A Verilog simulations based on physics data (128 channels, 5000 time steps) Good efficiency for top end of plot : later time stamps Poor efficiency at bottom end of plot: clusters have lower priority and can be overwritten

13. Readout system components X 16 6 bit chip output (5bit pixel data Plus header) Data present x16 Intermediate memory Column memory Column memory Column memory First level multiplexer x16 Intermediate memory Column memory Column memory Column memory First level multiplexer Top level multiplexer (driven at 8x front end frequency) Multiplexers are based on rotating pointers in shift registers. During operation pointer positions become randomised ensuring all parts of the CCD have equal chance of being read out.

14. Cluster processing Layout segment 40 microns x1.7mm State machine Logic Buffer registers Time-stamp registers Memory cells (38 words) Header memory (38 bits) 5 bit ADC output (encoded) Code converter 1 clock delay reg 5 bit adder Partial sum Partial sum of Next channel 6 bit adder Dig threshold Register (local) Dig comparator Memory controller Raw data from Code converter Header register (36 bits) 3 word timestampPixel data memory (shift register type, 38 words) Output to intermediate memory via first level multiplexer Logic or of header reg Initiates readout by mux Extract data signal (from lev 1 mux)

15. Cell : 1 2 3 4 5 6 7 8 9 10 Register cells on 4  m pitch Mirrored layouts : Nwells of adjacent cells are butted together, to minimise transistor separations Layout is 3 times more compact than CPR2 non-mirrored layout Sections of digital layout Layout of 16 columns: 320  m x 3.5 mm Code converter Logic: Adders Threshold store Comparators State machine (front end) State machine (back end) Layout of 2 columns : 40  m x 30  m

16. Area 320 x 315  m Provides local storage of data and header bits Interface between first and second level mux stages New compact layouts - 6 bits in separate blocks over 320 microns intermediate memory

17. Conclusions Status CPR2A chip has been designed to build on the success of the CPR2, but with greater memory depth to reduce dead time problems. CPR2A layout largely completed. Final verifications with realistic simulated physics data are being performed. Submission by October Next steps Plan to realise further versions of the readout chip on 0.13  m technology. Next version will have more memory and a more sophisticated sparsification algorithm to reduce dead time problem.