130nm Digital Sampler chip New results Jean-François Genat on behalf of D. Fougeron, 1 R. Hermel 1, H. Lebbolo 2, T.H. Pham 2, R. Sefri, and A. Savoy-Navarro 2 1 LAPP Annecy, 2 LPNHE Paris 5th SiLC Meeting, April 25-27th 2007, Prague 2 Outline • Goals • Last results • New results • Further tests • Next chip J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 Goals This chip Full readout chain integration in a single chip - Preamp-shaper Trigger decision (analog sums) Pulse Sampling: Analog pipe-lines On-chip digitization: ADC Buffering and pre-processing: Centroids, Least square fits, - Lossless compression and error codes - Calibration and calibration management - Power switching (ILC timing) • • • CMOS 130nm CMOS 90nm 512-1024 channels envisaged J-F Genat, 5th SiLC meeting, 2006 2007 Prague, April 25-27th 2007 yes yes yes yes yes no no no no yes no 4 channels Front-end in 130nm 130nm CMOS: Smaller Faster More radiation tolerant Lower power Will be (is) dominant in industry Drawbacks: - J-F Genat, 5th SiLC meeting, Reduced voltage swing (Electric field constant) Leaks (gate/subthreshold channel) Models more complex, not always up to date Crosstalk (digital-analog) Prague, April 25-27th 2007 2006 Chips Summary 130nm Chip #1 Under tests - Preamp-shapers + Sparsifier Pipeline 1 ADC Digital Chip #2 J-F Genat, 5th SiLC meeting, (4 channels) (One channel) Preamp-shapers + Sparsifier DC servo Pipeline 2 DAC Test structures: MOSFETS, passive Prague, April 25-27th 2007 4-channel chip Channel n+1 Sparsifier Can be used for a “trigger” S aiVi > th Time tag Channel n-1 reset reset Analog samplers, (slow) Wilkinson ADC Strip Ch # Preamp + Shapers Waveforms Counter UMC CMOS 130nm Clock 3-96 MHz J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 Targeted Amplifier: Shaper: Sparsifier: Sampler: ADC: 20 mV/MIP gain threshold on analog sum 16-deep 10-bit Noise: Measured with 180nm CMOS: 375 + 10.5 e-/pF @ 3 ms shaping, 90mW power J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 Outline • 130 nm chip goals • February results • Present results • Further tests • Next chip J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 Silicon 180nm 130nm Picture J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 130nm chip noise results (2/’07) Gain OK: Dynamics: Peaking time: 30 mV/MIP 30 MIPs @ 5% .8 – 2ms OK OK .7 - 3 ms Power (Preamp+ Shaper) = 290 mW Noise: 130nm @ 0.8 ms : 850 + 14 e-/pF 130nm @ 2 ms : 625 + 9 e-/pF 180nm @ 3 ms : 375 + 10.5 e-/pF J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 625*sqrt(2/3 ms)=510 e-/pF Outline • 130 nm chip goals • Last results • Present results • Further tests • Next chip J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 Analog pipeline output Simulation of the analog pipeline J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 Measured output of the ADC ADC J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 Sampler + ADC Waveform biased by the output pad parasitic capacitance (~ 1pF) Chip130-2 under tests: buffer between sampler and ADC J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 Outline • 130 nm chip goals • Last results • Present results • Further tests • Next chip J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 130-1 next tests Measure ADC extensively Linearities integral, differential Noise fixed pattern, random Speed maximum clock rate Effective number of bits (ENOB) J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 130nm Chip 2 DC Servo Sparsifier Channel n-1 S ai Vi > th ADC Channel n+1 ramp Analog Pipe-line Preamp and Shapers J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 130-2 tests (LAPP Annecy) Measure improved pipeline extensively Denis Fougeron’s (LAPP) design Linearities integral, differential Noise fixed pattern, random Speed maximum clock rate Droop Hold data for 1ms at ILC J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 Outline • 130 nm chip goals • Last results • Present results • Further tests • Next chip J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 Next chip Equip a detector: Lab test bench and 2007 beam-tests 130nm chips 1 + 2 128 channels - J-F Genat, 5th SiLC meeting, Preamp-shapers + sparsifier Pipeline ADC Digital Calibration Power cycling Prague, April 25-27th 2007 Power cycling Switch the current sources between zero and a small fraction (10-2 to 10-3 ) This option switches the current source feeding both the preamplifier & shaper between 2 values to be determined by simulation. Zero or a small fraction (0.1% - 1%) of biasing current is held during « power off ». Zero-power option tested on 180nm chip with 2 ms recover time constant J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 Planned Digital Front-End Chip control Buffer memory Processing for - Calibrations - Amplitude and time least squares estimation, centroids - Raw data lossless compression Tools - Digital libraries in 130nm CMOS available - Synthesis from VHDL/Verilog - SRAM - Some IPs: PLLs - Need for a mixed-mode simulator J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 Some issues with 130nm design Noise likely not model pessimistic, but measured acceptable Design rules more constraining Some (via densities) not available under Cadence Calibre (Mentor) required Lower power supplies voltages Low Vt transistors leaky (Low leakage option available) J-F Genat, 5th SiLC meeting, Prague, April 25-27th 2007 The End … Backup Jean-François Genat 3d SiLC Workshop, June 13-14th 2006, Liverpool Possible issues: noise: 130nm vs 180nm (simulation) PMOS: 130nm W/L = 2mm/0.5u Ids = 38.79u,Vgs=-190mV,Vds=-600mV gm=815.245u,gms=354.118u 1MHz 7.16nV/sqrt(Hz) Thermal noise hand calculation = 3.68nV/sqrt(Hz) Jean-François Genat 180nm gm=944.4uS,gms=203.1uS 1MHz 3.508nV/sqrt(Hz) Thermal noise hand calculation = 3.42nV/sqrt(Hz) 3d SiLC Workshop, June 13-14th 2006, Liverpool Noise: 130nm vs 180nm (simulation) NMOS : 130nm W/L = 50u/0.5u Ids=48.0505u,Vgs=260mV,Vds=1.2V gm=772.031uS,gms=245.341uS,gds=6.3575uS 180nm W/L=50u/0.5u Ids=47uA,Vgs=300mV,Vds=1.2V gm=842.8uS,gms=141.2uS,gds=16.05uS 1MHz --> 24.65nV/sqrt(Hz) 1MHz --> 4nV/sqrt(Hz) 100MHz --> 5nV/sqrt(Hz) Thermal noise hand calculation = 3.78nV/sqrt(Hz) 10MHz --> 3.49nV/sqrt(Hz) Thermal noise hand calculation = 3.62nV/sqrt(Hz) Jean-François Genat 3d SiLC Workshop, June 13-14th 2006, Liverpool Preamp CR RC Shaper Follower Comparator 130-90nm noise evaluation (STM process) L. Ratti et al FE2006 Perugia 1 MHz 100 mA J-F Genat 4th SiLC Workshop 1mA Dec 18th – 20th 2006 Barcelone Noise origin Thermal noise white 4 kT/gm Excess noise: not so white… due to process features (?) 1/f noise due to: - Traps at the gate dielectrics interfaces - Mobility fluctuations in the channel UMC models and measurements: - The gate dielectric in 130nm induces more traps compared to 180nm degrading 1/f noise. No more SiO2 in 90nm next generation CMOS -> hi-k dielectrics (Hf, Zr …) - Some thermal excess noise shows up Ratti et al. (Pavia, Bergamo), CERN: - Both ST90nm and IBM 130nm show better 1/f noise performance To be looked at : UMC 90nm simulations J-F Genat 4th SiLC Workshop Dec 18th – 20th 2006 Barcelone Backup Silicon strips readout using CMOS Deep Sub-Micron Technologies Jean-Francois Genat Leuven, Feb 26th 2007 J. David, 2 M. Dellhot, D. Fougeron, 1 R. Hermel 1, H. Lebbolo2 , T.H. Pham 2, F. Rossel 2, A. Savoy-Navarro 2, R. Sefri 2 , S. Vilalte 1 2 1 LAPP Annecy, 2 LPNHE Paris Help from IMEC-Leuven (Europractice) C. Das, E. Deumens, P Malisse Outline • Detector data • Technologies • Front-End Electronics • 180nm chip • 130nm chips • Future plans Silicon strips parameters 4-5 106 Silicon strips 10 - 60 cm long Thickness Inner : 150 -200mm Outer : -400mm Strip pitch 50 mm Inner : Outer : Double sided AC coupled Single sided DC coupled Silicon strips data at the ILC Pulse height: Cluster centroid to get a few µm position resolution Detector pulse analog sampling to get accurate charge estimation Time: Two scales: Coarse : 150-300 ns for BC identification, 80ns sampling Shaping time of the order of the microsecond Fine: nanosecond timing for the coordinate along the strip 10ns sampling Not to replace another layer or double sided Position estimation to a few cm using pulse reconstruction from samples Shaping time: 20-50 ns Coordinate along the strip SPICE L =28nH R =5 W Ci=500 fF 15 ns 90cm V = 6 107 m/s= c/5 Cs= 100 fF V 1 / LC 1 ns time resolution is 6 cm Measured Pulse Velocity Measured velocity: 5.5cm/ns Measured moving a laser diode along 24 cm Outline • Detector data • Technologies • Front-End Electronics • 180nm chip • 130nm chips • Future plans Technologies Silicon detector and VLSI technologies allow to improve detector and front-end electronics integration Front-end chips: Thinner CMOS processes 250, 180, 130, 90 nm now available from UMC, TSMC SiGe, less 1/f noise, faster Chip thinning down to 50 mm ? More channels on a chip, more functionalities, less power Connectivity: 3D ? On detector bump-bonding ? Stud bonding ? Stud bonding sufficient if pitch more than 50mm Smaller pitch detectors, better position and time resolution. Less material Outline • Detector data • Technologies • Front-End Electronics • 180nm chip • 130nm chips • Future plans Integrated functionalities Full readout chain integration in a single chip - Preamp-shaper - Trigger decision (analog sums) : compact data - Sampling: Analog pipe-lines - On-chip digitization - Buffering - Digital Processing: Centroids and Least Squares time/amplitude estimation - Calibration and calibration management - Power switching Presently 128 channels (APV, SVX) chips, 256-1024 envisaged Front-End Chip Integrate 512-1024 channels in 90nm CMOS: Charge Amplifiers: 20-30 mV/MIP over 30 MIP Shapers: - slow 500 ns – 1 ms - fast 20-50 ns Zero-suppression: threshold the sum of adjacent channels 2D analog memory: - 8-16 samples 80 ns and 10 ns sampling clocks - Event buffer 16-deep ILC timing: ADC: 10 bits Buffering Calibration Power switching saves a factor 100-200 1 ms: ~ 3-6000 trains @150-300ns / BC 200ms in between Foreseen Front-end architecture Channel n+1 ‘trigger’ Sparsifier S aiVi > th Time tag Channel n-1 reset Strip Wilkinson ADC Calibration Control reset Analog samplers, slow, fast Ch # Waveforms Preamp + Shapers Counter Charge 1-40 MIP, Time resolution: BC tagging 150-300ns, Technologies: Future: Storage Deep Sub-Micron CMOS 180-130nm SiGe &/or deeper DSM fine: ~ 1ns Outline • Detector data • Technologies • Front-End Electronics • 180nm chip • 130nm chips • Future plans J-F Genat, LECC06, Valencia, Sept 25-29th 2006 Silicon 2005 - Preamp - Shaper - Sample & Hold - Comparator 16 + 1 channel UMC 180nm chip 3mm (layout and picture) Process spreads Process spreads: 3.3 % (same wafer) Preamp gains statistics (same wafer) Shaper output noise 375 e- RMS 375 e- +10.4 e-/pF input noise with chip-on-board wiring 275 + 8.9/pFsimulated Linearities 180nm chip +/-1.5% +/-0.5% expected +/-6% +/-1.5% expected Tests conclusions on UMC 180 nm chips 12 chips tested (end 2005) The UMC CMOS 180nm process is mature and reliable: - Models mainly OK - Only one transistor failure over 12 chips - Process spreads of a few % 90Sr Source tests with 180nm chip connected to a Si detector and compared with VA1 chip from Ideas (see test bench section) S/N ~ 10 still system noise dominating Outline • Detector data • Technologies • Front-End Electronics • 180nm chip • 130nm chips • Future plans Front-end in CMOS 130nm 2006-2007 130nm CMOS motivations: Benefits - Smaller Faster More radiation tolerant Less power Presently dominant in the IC industry Issues: - Design more constraining (Design rules) Reduced voltage swing (Electric field constant) Leaks (gate/subthreshold channel) Models more complex, sometimes still not accurate UMC Technology parameters • • • • • 3.3V transistors Logic supply Metals layers MIM capacitors Transistors 180 nm 130nm yes 1.8V 6 Al 1fF/mm² Three Vt options yes 1.2V 8 Cu 1.5 fF/mm2 Low leakage option Useful for analog storage during < 1 ms 130nm 4-channel test chip Channel n+1 Zero-suppression Can be used for a “trigger” S aiVi > th Time tag Channel n-1 reset reset Analog samplers, (slow) Ramp ADC Strip Ch # Preamp + Shaper DC servo implemented for DC coupled detectors UMC CMOS 130nm Received in 2006 Being tested: Analog OK, Digital under tests Waveforms Counter Clock 3-96 MHz Analog pipeline simulation Silicon 180nm 130nm Picture Noise 130nm vs 180nm (simulation) PMOS: 180nm 130nm gm=944.4uS Id = 30 uA Weak inversion 1MHz 3.508nV/sqrt(Hz) gm=815.245uS Id =60 uA Moderate inversion 1MHz 7.16nV/sqrt(Hz) Thermal noise hand calculation = 3.42nV/sqrt(Hz) Thermal noise hand calculation = 3.68nV/sqrt(Hz) Measurements on 130nm show 2 x better results ! 130nm-1 design Noise models pessimistic ? 130nm CMOS Silicon actually 2x better ! Design rules more constraining Post layout simulations at Leuven (Mentor) Some design rules (via densities) not available under Cadence Calibre (Mentor) required 130nm-1 conclusions Good matching wrt simulation Except noise which is… much better ! - Still lot to test: pipe-line and digital - Statistics available from two wafers (70 chips) Very encouraging results Go to 90 nm ? 130nm-2 design One channel version with - Servo DC - Improved pipe-line - Calibration DAC Layout Picture One channel 1.5 x 1.5 mm2 130nm-2 architecture DC servo to accommodate DC coupled detectors, DAC + calibration, Improved pipe-line Preamp Shaper DC reference Received January 5th 2007 Test card under wiring Test stand under work Analog sampler Planned on-chip digital Chip control Buffer memory Processing for - Calibrations - Amplitude and time least squares estimation, centroids Tools - Digital libraries in 130nm CMOS available (VST) - Place & Route tools: Cadence + design kits from Europractice (Leuven, Belgium) Mixed mode needed - Synthesis from VHDL/Verilog - Some IPs: PLLs, SRAM Wiring Detector to FE Chips Wire bonding Flip Chip Technology Courtesy: Marty Breidenbach (Cal SiD) OR (later) Wiring Detector to FE Chips Courtesy: Ray Yarema, FEE 2006, Perugia 3D Wiring Courtesy: Ray Yarema, FEE 2006, Perugia Manuel Lozano (CNM Barcelona) Chip connection • Wire bonding – Only periphery of chip available for IO connections – Mechanical bonding of one pin at a time (sequential) – Cooling from back of chip – High inductance (~1nH) – Mechanical breakage risk (i.e. CMS, CDF) • Flip-chip – Whole chip area available for IO connections – Automatic alignment – One step process (parallel) – Cooling via balls (front) and back if required – Thermal matching between chip and substrate required – Low inductance (~0.1nH) J-F Genat, LECC06, Valencia, Sept 25-29th 2006 Manuel Lozano (CNM Barcelona) Bump bonding flip chip technology • Electrical connection of chip to substrate or chip to chip face to face flip chip • Use of small metal bumps bump bonding CNM • Process steps: – Pad metal conditioning: Under Bump Metallisation (UBM) – Bump growing in one or two of the elements – Flip chip and alignment – Reflow – Optionally underfilling Manuel Lozano (CNM Barcelona) Bump bonding flip chip technology • Bumping technologies – Evaporation through metallic • Expensive technology mask – Especially for small quantities – Evaporation with thick (as in HEP) photoresist – Big overhead of NRE costs – Screen printing • Minimal pitch reported: 18 µm but ... – Stud bumping (SBB) • Few commercial companies for fine – Electroplating pitch applications (< 75 µm) – Electroless plating – Conductive Polymer Bumps – Indium evaporation The End … The SiTR-130_1 chip 180nm 130nm Picture Possible issues: noise: 130nm vs 180nm (simulation) PMOS: 180nm 130nm gm=944.4uS 1MHz 3.508nV/sqrt(Hz) gm=815.245uS 1MHz 7.16nV/sqrt(Hz) Thermal noise hand calculation = 3.42nV/sqrt(Hz) Thermal noise hand calculation = 3.68nV/sqrt(Hz) Measurements show 2 times better results SiTR-130_1 tests results Gain OK: 30 mV/MIP Dynamics: 30 MIPs @ 5% Peaking time: 0.8 – 2ms OK OK 0.7 - 3 ms Power (Preamp+ Shaper) = 300 mW Noise comparative 130nm @ 0.8 ms : 850 + 14e-/pF 130nm @ 2 ms : 625 + 9e-/pF 180nm @ 3 ms : 360 + 10.5 e-/pF Some issues with 130nm design Noise models pessimistic, Silicon actually much better ! Design rules more constraining Some design kits are not fully developed and thus additional effort needed Power dissipation budget (measured) Preamp Shaper 180nm/ch 90 180 130nm/ch 148 148 Common Zero suppr. Pipeline Total Analog ADC Logic Total Digital 270 198 10 100 575 66 5 96 101 Final goal: ≤ 1 mWatt/channel all included (looks achievable) Next developments Implement the fast (20-50ns shaping) version in Silicon-Germanium including: - Preamp + Shaper (20-100ns) - Fast sampling Submit a full 128 channel version in 130nm CMOS including slow and fast analog processing, power cycling, digital Planned on-chip digital Chip control Buffer memory Processing for - Calibrations - Amplitude and time least squares estimation, centroids Tools - Digital libraries in 130nm CMOS available (VST) - Place & Route tools: Cadence + design kits from Europractice (Leuven, Belgium) - Synthesis from VHDL/Verilog - Some IPs: PLLs, SRAM Chip connection on µstrips • Wire bonding – Only periphery of chip available for IO connections – Mechanical bonding of one pin at a time (sequential) – Cooling from back of chip – High inductance (~1nH) – Mechanical breakage risk • Flip-chip – Whole chip area available for IO connections – Automatic alignment – One step process (parallel) – Cooling via balls (front) and back if required – Thermal matching between chip and substrate required – Low inductance (~0.1nH) Next developments Implement the fast (20-50ns shaping) version in Silicon-Germanium including: - Preamp + Shaper (20-100ns) - Fast sampling Submit a full 128 channel version in 130nm CMOS including slow and fast analog processing, power cycling, digital Planned on-chip digital Chip control Buffer memory Processing for - Calibrations - Amplitude and time least squares estimation, centroids Tools - Digital libraries in 130nm CMOS available (VST) - Place & Route tools: Cadence + design kits from Europractice (Leuven, Belgium) - Synthesis from VHDL/Verilog - Some IPs: PLLs, SRAM Chip connection on µstrips • Wire bonding – Only periphery of chip available for IO connections – Mechanical bonding of one pin at a time (sequential) – Cooling from back of chip – High inductance (~1nH) – Mechanical breakage risk • Flip-chip – Whole chip area available for IO connections – Automatic alignment – One step process (parallel) – Cooling via balls (front) and back if required – Thermal matching between chip and substrate required – Low inductance (~0.1nH) 1st step: bump bonding FE on detector Studying flip-chip FEE chips on strips IMB-CNM, VTT, HIP, LPNHE, Liverpool (firms) • Design of the fan-in fan-out on the µstrip sensor 1st case: 128 ch 2nd case: 1024 channel The design is starting have a foundry or a firm ready to implement it on the sensor • Bump bonding of the FEE chip crucial issue here: have a working chip with the required nb of channels (by end 2007) 2nd step: cabling => use: Tape Automated Bonding (TAB) or???????? Don’t forget auxiliary electronics that degrade any fancy solution on paper ex: decoupling capacitors A few words about cabling and connection to the overall DAQ system: preliminary thoughts • A preparer… Chip connection on µstrips • Wire bonding – Only periphery of chip available for IO connections – Mechanical bonding of one pin at a time (sequential) – Cooling from back of chip – High inductance (~1nH) – Mechanical breakage risk • Flip-chip – Whole chip area available for IO connections – Automatic alignment – One step process (parallel) – Cooling via balls (front) and back if required – Thermal matching between chip and substrate required – Low inductance (~0.1nH) 1st step: bump bonding FE on detector Studying flip-chip FEE chips on strips IMB-CNM, VTT, HIP, LPNHE, Liverpool (firms) • Design of the fan-in fan-out on the µstrip sensor 1st case: 128 ch 2nd case: 1024 channel The design is starting have a foundry or a firm ready to implement it on the sensor • Bump bonding of the FEE chip crucial issue here: have a working chip with the required nb of channels (by end 2007) 2nd step: cabling => use: Tape Automated Bonding (TAB) or???????? Don’t forget auxiliary electronics that degrade any fancy solution on paper ex: decoupling capacitors Next developments Implement the fast (20-50ns shaping) version in Silicon-Germanium including: - Preamp + Shaper (20-100ns) - Fast sampling Submit a full 128 channel version in 130nm CMOS including slow and fast analog processing, power cycling, digital Planned on-chip digital Chip control Buffer memory Processing for - Calibrations - Amplitude and time least squares estimation, centroids Tools - Digital libraries in 130nm CMOS available (VST) - Place & Route tools: Cadence + design kits from Europractice (Leuven, Belgium) - Synthesis from VHDL/Verilog - Some IPs: PLLs, SRAM Chip connection on µstrips • Wire bonding – Only periphery of chip available for IO connections – Mechanical bonding of one pin at a time (sequential) – Cooling from back of chip – High inductance (~1nH) – Mechanical breakage risk • Flip-chip – Whole chip area available for IO connections – Automatic alignment – One step process (parallel) – Cooling via balls (front) and back if required – Thermal matching between chip and substrate required – Low inductance (~0.1nH) 1st step: bump bonding FE on detector Studying flip-chip FEE chips on strips IMB-CNM, VTT, HIP, LPNHE, Liverpool (firms) • Design of the fan-in fan-out on the µstrip sensor 1st case: 128 ch 2nd case: 1024 channel The design is starting have a foundry or a firm ready to implement it on the sensor • Bump bonding of the FEE chip crucial issue here: have a working chip with the required nb of channels (by end 2007) 2nd step: cabling => use: Tape Automated Bonding (TAB) or???????? Don’t forget auxiliary electronics that degrade any fancy solution on paper ex: decoupling capacitors backup J-F Genat, LECC06, Valencia, Sept 25-29th 2006 Beam-tests at DESY October 2006 J-F Genat, LECC06, Valencia, Sept 25-29th 2006 Wiring Detector to FE Chips Wire bonding Flip Chip Technology Courtesy: Marty Breidenbach (Cal SiD) OR (later) J-F Genat, LECC06, Valencia, Sept 25-29th 2006 Wiring Detector to FE Chips Courtesy: Ray Yarema, FEE 2006, Perugia J-F Genat, LECC06, Valencia, Sept 25-29th 2006 3D Wiring Courtesy: Ray Yarema, FEE 2006, Perugia J-F Genat, LECC06, Valencia, Sept 25-29th 2006 Manuel Lozano (CNM Barcelona) Chip connection • Wire bonding – Only periphery of chip available for IO connections – Mechanical bonding of one pin at a time (sequential) – Cooling from back of chip – High inductance (~1nH) – Mechanical breakage risk (i.e. CMS, CDF) • Flip-chip – Whole chip area available for IO connections – Automatic alignment – One step process (parallel) – Cooling via balls (front) and back if required – Thermal matching between chip and substrate required – Low inductance (~0.1nH) J-F Genat, LECC06, Valencia, Sept 25-29th 2006 Manuel Lozano (CNM Barcelona) Bump bonding flip chip technology • Electrical connection of chip to substrate or chip to chip face to face flip chip • Use of small metal bumps bump bonding CNM • Process steps: – Pad metal conditioning: Under Bump Metallisation (UBM) – Bump growing in one or two of the elements – Flip chip and alignment – Reflow – Optionally underfilling J-F Genat, LECC06, Valencia, Sept 25-29th 2006 Manuel Lozano (CNM Barcelona) Bump bonding flip chip technology • Bumping technologies – Evaporation through metallic • Expensive technology mask – Especially for small quantities – Evaporation with thick (as in HEP) photoresist – Big overhead of NRE costs – Screen printing • Minimal pitch reported: 18 µm but ... – Stud bumping (SBB) • Few commercial companies for fine – Electroplating pitch applications (< 75 µm) – Electroless plating – Conductive Polymer Bumps – Indium evaporation J-F Genat, LECC06, Valencia, Sept 25-29th 2006 Noise: 130nm vs 180nm (simulation) NMOS : 130nm W/L = 50u/0.5u Ids=48.0505u,Vgs=260mV,Vds=1.2V gm=772.031uS,gms=245.341uS,gds=6.3575uS 180nm W/L=50u/0.5u Ids=47uA,Vgs=300mV,Vds=1.2V gm=842.8uS,gms=141.2uS,gds=16.05uS 1MHz --> 24.65nV/sqrt(Hz) 1MHz --> 4nV/sqrt(Hz) 100MHz --> 5nV/sqrt(Hz) Thermal noise hand calculation = 3.78nV/sqrt(Hz) 10MHz --> 3.49nV/sqrt(Hz) Thermal noise hand calculation = 3.62nV/sqrt(Hz) J-F Genat, LECC06, Valencia, Sept 25-29th 2006 Noise: 130nm vs 180nm (UMC) Models 1/f NMOS Thermal 130/180: factor 6 worse 130/180: factor 1.4 worse 130nm W/L = 50/0.5 Ids=48 uA gm=770uS 180nm W/L=50/0.5 Ids=47uA gm=840uS 1MHz --> 24 nV/sqrt(Hz) 1MHz --> 4 nV/sqrt(Hz) 100MHz --> 5nV/sqrt(Hz) Calculation = 3.8nV/sqrt(Hz) 10MHz --> 3.5nV/sqrt(Hz) Calcualtion= 3.6nV/sqrt(Hz) Measurements give an overall factor of 1.5 between 130 and 180nm at 800 ns and 3ms shaping time (look at 130nm at 2 ms) J-F Genat 4th SiLC Workshop Dec 18th – 20th 2006 Barcelone Manuel Lozano (CNM Barcelona) Chip connection • Wire bonding – Only periphery of chip available for IO connections – Mechanical bonding of one pin at a time (sequential) – Cooling from back of chip – High inductance (~1nH) – Mechanical breakage risk (i.e. CMS, CDF) • Flip-chip – Whole chip area available for IO connections – Automatic alignment – One step process (parallel) – Cooling via balls (front) and back if required – Thermal matching between chip and substrate required – Low inductance (~0.1nH) Manuel Lozano (CNM Barcelona) Bump bonding flip chip technology • Electrical connection of chip to substrate or chip to chip face to face flip chip • Use of small metal bumps bump bonding CNM • Process steps: – Pad metal conditioning: Under Bump Metallisation (UBM) – Bump growing in one or two of the elements – Flip chip and alignment – Reflow – Optionally underfilling Manuel Lozano (CNM Barcelona) Bump bonding flip chip technology • Bumping technologies – Evaporation through metallic • Expensive technology mask – Especially for small quantities – Evaporation with thick (as in HEP) photoresist – Big overhead of NRE costs – Screen printing • Minimal pitch reported: 18 µm but ... – Stud bumping (SBB) • Few commercial companies for fine – Electroplating pitch applications (< 75 µm) – Electroless plating – Conductive Polymer Bumps – Indium evaporation 0.18 mm chip PREAMPLIFIER Reset FET Gain: 8mV/MIP 3.3V input trans 2000/0.5 gm = 0.69 mS 40 mA (Weak inversion IC ~= 0.01) Tests results : Gain OK Linearity : +/-1.5% +/-0.5% expected Noise 3ms-20ms rise-fall, 40 mA : 498 + 16.5 e-/pF OK Dynamic range : 60 MIPs OK Jean-François Genat 3d SiLC Workshop, June 13-14th 2006, Liverpool 0.18 mm chip SHAPER RC-CR Peaking time ajustable :1ms -> 5ms Tests: 1.5 - 6 ms rise/fall Linearity: +/- 6% +/- 1% expected Noise @ 3 ms 140mW power : 375 + 10.4 e-/pF 274 + 8.9 e-/pF expected Jean-François Genat 3d SiLC Workshop, June 13-14th 2006, Liverpool Circuit en 0.13 mm PREAMPLIFIER Gain: 28.5mV/MIP 3.3V transistor d’entrée 1.5m/0.5mm gm = 1.5 mS/ ~70 mA Capacité de charge = 133 fF Plage de dynamique : 17MIPs (linearité = 1% a 17MIPs) Bruit@70 mA = 1000e- + 25e-/pF SHAPER Filtre RC-CR Temps adjusté :0.7ms -> 3ms bruit@700ns = 335+22e-/pF bruit@3ms = 305+16e-/pF Plage de dynamique : 17MIPS Jean-François Genat 3d SiLC Workshop, June 13-14th 2006, Liverpool 0.13 mm Chip Sparsifier SPARSIFICATION + Sum of3 adjacent channels + Résolution ~ 0.1mV + Response time = 186ns Sommateur Offset cancel Ref Jean-François Genat Comparateur 3d SiLC Workshop, June 13-14th 2006, Liverpool 0.13 mm Chip Simulations SIMULATIONS : Preamps linearity Shaper linearity Sparsifier response Jean-François Genat 3d SiLC Workshop, June 13-14th 2006, Liverpool Possible issues: Transistors leaks Two situations: - Gate-channel due to tunnel effect (can affect noise performances) - Through channel when transistor switched-off (only affects large digital designs) Sub-threshold current Nano-CMOS Circuit and Physical design B.P Wong, A. Mittal, Y. Cao, G. Starr, 2005, Wiley 130 nm 180 nm Gate leakage Jean-François Genat 90 nm - 180nm chip OK - 130nm, no gate leakage expected, but sub-threshold - 90nm, important gate leakage Scale: 1 nA/mm = 8000 e- noise in FE 3d SiLC Workshop, June 13-14th 2006, Liverpool Possible issues: noise: 130nm vs 180nm (simulation) PMOS: 130nm W/L = 2mm/0.5u Ids = 38.79u,Vgs=-190mV,Vds=-600mV gm=815.245u,gms=354.118u 1MHz 7.16nV/sqrt(Hz) Thermal noise hand calculation = 3.68nV/sqrt(Hz) Jean-François Genat 180nm gm=944.4uS,gms=203.1uS 1MHz 3.508nV/sqrt(Hz) Thermal noise hand calculation = 3.42nV/sqrt(Hz) 3d SiLC Workshop, June 13-14th 2006, Liverpool Noise: 130nm vs 180nm (simulation) NMOS : 130nm W/L = 50u/0.5u Ids=48.0505u,Vgs=260mV,Vds=1.2V gm=772.031uS,gms=245.341uS,gds=6.3575uS 180nm W/L=50u/0.5u Ids=47uA,Vgs=300mV,Vds=1.2V gm=842.8uS,gms=141.2uS,gds=16.05uS 1MHz --> 24.65nV/sqrt(Hz) 1MHz --> 4nV/sqrt(Hz) 100MHz --> 5nV/sqrt(Hz) Thermal noise hand calculation = 3.78nV/sqrt(Hz) 10MHz --> 3.49nV/sqrt(Hz) Thermal noise hand calculation = 3.62nV/sqrt(Hz) Jean-François Genat 3d SiLC Workshop, June 13-14th 2006, Liverpool