Serial Powering vs. DC-DC Conversion - A First Comparison Tracker Upgrade Power WG Meeting October 7 th, 2008 Katja Klein 1. Physikalisches Institut B RWTH Aachen University
Outline Compare Serial Powering & DC-DC conversion under various aspects Power loss in cables Local efficiency Compatibility with services Power supplies Bias voltage Safety Slow control Start-up Scalability Flexibility Potential to deliver different voltages Process considerations & radiation hardness Interplay with FE-chip Interplay with readout & controls Noise Material budget Space Test systems Discussion Katja Klein Serial Powering vs. DC-DC Conversion 2
The Basic Ideas Powered from constant current source Each module is on different ground potential AC-coupled communication Shunt regulator and transistor to take excess current and stabilize voltage Voltages are created locally via shunt and linear regulators Parallel powering with DC-DC conversion Need radiation-hard magnetic field tolerant DC-DC converter One converter per module or parallel scheme 1-step or 2-step conversion V drop = R I 0 P drop = R I 0 2 Conversion ratio r: r = V out / V in! << 1 P drop = R I 02 n 2 r 2 Katja Klein Serial Powering vs. DC-DC Conversion 3
The Buck Converter The buck converter is simplest inductor-based step-down converter: Convertion ratio g > 1: g = V in / V out Switching frequency f s : f s = 1 / T s Katja Klein Serial Powering vs. DC-DC Conversion 4
The Charge Pump Capacitor-based design Step-down: capacitors charged in series and discharged in parallel Conversion ration = 1 / number of parallel capacitors Low currents Katja Klein Serial Powering vs. DC-DC Conversion 5
Implementation Examples : : Atlas pixels, Tobias Stockmanns Regulators on-chip or on the hybrid AC-coupled communication with off-module electronics Power for optical links not integrated HV not integrated Stefano Michelis, TWEPP2008 Two-stage system Diff. technologies proposed for the two stages Analogue and digital power fully separated Power for optical links ~ integrated HV not integrated Katja Klein Serial Powering vs. DC-DC Conversion 6
What Conversion Ratio do we need? Conversion ratio needed for parallel powering with DC-DC converters? Total tracker current estimate Current strip tracker: 15kA; current pixel: 1.5kA Geoffs strawman: strips: 25kW/1.2V = 21kA; pixels: 3.2kA; trigger layers: 10kA Currents increase roughly by factor of 2 in this strawman Power loss in cables Goes with I 2 increase by factor of 4 for same number of cables (2000) Total power loss inverse proportional to number of power groups Can compensate with (conversion ratio) 2 Material budget Saving in cable x-section scales with I Total material independent of segmentation Of course want to reduce as much as possible With conversion ratio of ¼ we would be as good as or better than today. SP: current fixed; cable material & power loss depends only on # of cables! Katja Klein Serial Powering vs. DC-DC Conversion 7
Power Losses in Cables Power losses in cables lead to decrease of overall power efficiency expensive... increase the heat load within the cold volume cooling capacity must be higher SP DC-DC, r = 1/5 DC-DC, r = 1/10 Consider system with n modules: P det = n I 0 V 0 Voltage drop on cables & power loss P cable calculated within each scheme Efficiency = P det / P total = P det / (P det + P cable ) Eff. increases with n. Since 10-20 modules can be chained, efficiency can be very high! Eff. goes down with n. Need more cables or lower conversion ratio Equal to SP if conversion ratio = 1/n Katja Klein Serial Powering vs. DC-DC Conversion 8
Local Efficiency Constant current source total power consumption is contant! Current of chain is fixed to highest current needed by any member Current not used by a module flows through shunt regulator Linear regulator: voltage difference between dig. & analog drops across it Local power consumption is increased! Estimated increase for - Atlas pixels (NIM A557): 35% - Atlas strips (NIM A579, ABCD): 18% All DC-DC converters have inefficiencies switching losses ESR of passive components R on of transistor etc. Typical values (e.g. comm. buck): 80-95% Efficiency goes down for low conv. ratio! Trade-off betw. eff. & switching frequency In two-step schemes, efficiencies multiply Estimates (St. Michelis, TWEPP2008): Step-1: 85-90% Step-2: 93% Total: 80-85% This needs to be demonstrated Katja Klein Serial Powering vs. DC-DC Conversion 9
Compatibility with LIC Cables Constraints from recycling of current services: 2000 LICs with two LV conductors & common return each Not realistic to split return to obtain 4000 lines Stay with 2000 LV power lines ( power groups ) LV conductors certified for 30V and 20A Twisted pairs (HV/T/H/sense) certified for 600V 256 PLCC control power cables Adapt at PP1 to (lower mass) cables inside tracker Current is small 30V allows for chains with more than 20 modules looks compatible 30V is largely enough For any reasonable segmentation and conv. factor currents should be lower e.g. 20 chips a 53mA per module 1.2A / module 20 modules per rod 24A /rod r = ¼ I = 6A looks compatible Katja Klein Serial Powering vs. DC-DC Conversion 10
Power Supplies Assume that power supplies will be exchanged after 10 years Constant current source Not so common in industry (e.g. CAEN) Atlas: PSs developed by Prague group (developed already their current PSs) No sensing Standard PS: ~15V, ~10A (radiation & magnetic field tolerant?) Any sensitivity of converter to input voltage ripple? No sensing needed (local regulation)? Katja Klein Serial Powering vs. DC-DC Conversion 11
Bias Voltage Power is not a problem (currents are very low) Up to now: independent bias lines for 1-2 modules Might not be possible anymore when current cables are re-used Note: T/H/sense wires are equal to HV wires Not yet well integrated into concept Derive on-module via step-up converters? In Atlas, piezo-electric transformers are discussed. Or independent delivery using todays cables Same options as for SP Katja Klein Serial Powering vs. DC-DC Conversion 12
Safety (I) Open leads to loss of whole chain Shunt regulators/transistors to cope with this Several concepts are on the market (next page) Connection to module can break bypass transistor on mothercable - high V, high I rad.-hardness? - must be controlable from outside Real-time over-current protection? Real time over-voltage protection? Open connections Converter itself can break Shorts between converter and module If PP of several mod.s by one converter: risk to loose several modules at once Fermilab expressed interest to perform a systematic failure analysis Katja Klein Serial Powering vs. DC-DC Conversion 13
Safety (II) 1. Shunt regulators + transistors parallel on-chip (Atlas pixels) + redundancy - matching issue at start-up Regulator with lowest threshold voltage conducts first all current goes through this regulator spread in threshold voltage and internal resistance must be small 2. One shunt regulator + transistor per module + no matching issue - no redundany - needs high-current shunt transistor - must stand total voltage 3. One reg. per module + distributed transistors + no matching issue + some redundancy - feedback more challenging Katja Klein Serial Powering vs. DC-DC Conversion 14
Slow Control Module voltage(s) Module current(s)? Bias current Slow control IC or block on hybrid Could be used to communicate with linear regulator and turn to stand-by Ideas to sense module voltage in Atlas pixels: - sense potential through HV return - sense through AC-coupled data-out termination - sense from bypass transistor gate Slow control IC or block on hybrid For on-chip charge pump: would be useful to have SC information from individual chips Could be used to set converter output voltage and switch on/off converters Katja Klein Serial Powering vs. DC-DC Conversion 15
Start-up & Selective Powering If controls powered from separate line, it can be switched on first Devices in chain switched on together (both module controller and FE-chips) Can take out modules only by closing bypass transistor from outside If converter output can be switched on/off, then easy and flexible: - controls can be switched on first - bad modules (chips?) can be switched off - groups of chips/modules can be switched on/off for tests This should be a requirement! Katja Klein Serial Powering vs. DC-DC Conversion 16
Scalability Consequences if more modules are powered per chain or in parallel? E.g. barrel vs. end caps: different # of modules per substructure Current is independent on # of modules Number of modules reflected in maximal voltage within chain; relevant for capacitors for AC-coupling constant current source bypass / shunt transistors If one converter per module: perfect scalability PP of several mod. by one converter: current depends on # of modules, must be able to power largest group Should specify soon what we need current per chip # of chips per module # of modules per substructure Otherwise we will be constraint by currents that devices can provide Katja Klein Serial Powering vs. DC-DC Conversion 17
Flexibility Flexibility with respect to combination of devices with different currents E.g. trigger vs. standard module (or 4 / 6-chips) Current of chain is equal to highest current needed by any member chains with mixed current requirements are inefficient! If one converter per module: very flexible, do not care! If PP of several modules by one converter: distribution between modules arbitrary Katja Klein Serial Powering vs. DC-DC Conversion 18
Potential to Provide Different Voltages Chip supply voltage(es): ~ 1.2V (Atlas: 0.9V for digital part to save power) Opto-electronics supply voltage: 2.5 3V Needed voltage created by regulators ~1.2V by shunt regulator Lower voltage derived from this via linear regulator efficiency loss Technically could power opto-electronics and controls via own regulators, but inefficient to chain devices with different current consumption Decouple from chain (Atlas: plan to power separately from dedicated cables) With charge pumps, only integer conversion ratios are possible With inductor-based designs, arbitrary V out < V in can be configured (but feedback circuit optimized for a certain range) Only hard requirement: V in >= V opto Analogue and digital voltage can be supplied independently no efficiency loss Katja Klein Serial Powering vs. DC-DC Conversion 19
Process Considerations & Radiation Hardness Regulators must be rad.-hard Standard CMOS process can be used; but... HV tolerant components (up to n U 0 ): - capacitors for AC-coupling - bypass transistor Shunt transistors must stand high currents (~2A) if one per module Commercial devices are not rad.-hard Apparent exception: Enpirion EN5360 (S. Dhawan, TWEPP2008) Standard 130nm CMOS: 3.3V maximal For high conversion ratio transistors must tolerate high V in, e.g. 12V Several high voltage processes exist Rad.-hard HV process not yet identified This is a potential show stopper For r = ½ (e.g. charge pump) can use 3.3V transistors - radiation hardness? Katja Klein Serial Powering vs. DC-DC Conversion 20
Interplay with FE-Chip Several options for shunt - Regulator and transistor on-chip - Only shunt transistor on-chip - Both external Linear regulators typically on-chip Next Atlas strip FE-chip (ABCnext): - linear regulator - shunt regulator circuit - shunt transistor circuit Next Atlas pixel chip (FE-I4): - Shunt regulator - LDO DC-balanced protocol Ideally fully decoupled Not true anymore in two-step approach with on-chip charge pump Next Atlas strip FE-chip (ABCnext): - linear regulator to filter switching noise Next Atlas pixel chip (FE-I4): - LDO - Charge pump (r = ½) No influence on protocol Katja Klein Serial Powering vs. DC-DC Conversion 21
Readout & Controls Modules are on different potentials AC-coupling to off-module electronics needed Decoupling either on the hybrid (needs space for chips & capacitors) or at the end of the rod (Atlas strips, P. Phillips, TWEPP08) Needs DC-balanced protocol increase of data volume Nothing special: electrical transmission of data and communication signals to control ICs No DC-balanced protocol needed Atlas pixels, NIM A557 Katja Klein Serial Powering vs. DC-DC Conversion 22
Noise Intrinsically clean - current is kept constant - voltages generated locally Main concerns: - pick-up from external source - pick-up from noisy module in chain Tests by Atlas pixels (digital) and strips (binary) revealed no serious problems - noise injection - modules left unbiased - decreased detection thresholds - external switchable load in parallel to one module (changes potential for all modules): some effect (Atlas pixels, NIM A557) Switching noise couples conductively into FE Radiated noise (actually magnetic near-field) is picked up by modules Details depend on FE, distances, filtering, coil type & design, switching frequency, conversion ratio,... Shielding helps against radiated noise, but adds material, work and cost LDO helps against conductive noise, but reduces efficiency Surprises might come with bigger systems Not good to start already with shielding and system-specific fine-tuning Katja Klein Serial Powering vs. DC-DC Conversion 23
Material Budget Regulators ~ one add. chip per hybrid Components for AC-coupling - HV-safe capacitors (might be big!) - LVDS chip Flex for discrete components Cable cross-section from PP1 to detector (rest stays) scales with current - One cable must carry I 0 - Total mass depends on modules / cable Motherboard/-cable: power planes can be narrow, small currents & voltages created locally Converter chip(s) Discrete components - air-core inductor (D = 1-2cm!) - output filter capacitor(s) Flex for discrete components One cable must carry I 0 nr total mass depends only on conv. ratio Motherboard/-cable - buck converter can tolerate certain voltage drop since input voltage must not be exact low mass - charge pumps have no output regulation: need exact V in Shielding? Katja Klein Serial Powering vs. DC-DC Conversion 24
Space Different options are discussed, but regulators + shunt transistors are either in readout chip or in a separate chip ~ one additional chip per hybrid Components for AC-coupling - LVDS buffers - HV-safe capacitors (might be big) Bypass transistor? Charge pump in readout chip or in a separate chip Buck converter: - controller chip - discrete air-core inductor (D = 1-2cm!) - discrete output filter capacitor(s) - more? very unlikely to be ever fully on-chip In all other inductor-based topologies more components (inductors!) needed Katja Klein Serial Powering vs. DC-DC Conversion 25
Test Systems for Construction Phase If AC-coupling at end of stave, a decoupling board is necessary to read out single modules Adapter PCB needed anyway for electrical readout Electrical readout of single modules possible with adapter PCB Katja Klein Serial Powering vs. DC-DC Conversion 26
Work on Powering within CMS Tracker RWTH Aachen (L. Feld) proposal accepted System test measurements with commercial and custom DC-DC (buck) converters Simulation of material budget of powering schemes Rad.-hard magnetic-field tolerant buck converter in collaboration with CERN group Bristol university (C. Hill) proposal accepted Development of PCB air-core toroid DC-DC converter designs with air-core transformer PSI (R. Horisberger) no proposal, but private communication Development of on-chip CMOS step-down converter (charge pump) IEKP Karlsruhe (W. de Boer) Powering via cooling pipes Fermilab / Iowa / Mississippi (S. Kwan) proposal under review proposal under review System test measurements focused on pixel modules (DC-DC conversion & SP) Power distribution simulation software Katja Klein Serial Powering vs. DC-DC Conversion 27
Both schemes have their pros and cons how to weigh them? SP is complicated, but I do not see a real show stopper DC-DC conversion is straightforward, but two potential show stoppers noise, radiation-hardness of HV-tolerant process Need to understand SP better In particular safety, slow controls Summary Up to now, we focus on DC-DC conversion should we start on SP? Who? Both Atlas pixels and strips integrate power circuitry in their new FE-chips: shunt regulators, charge pump, LDO Seems to be a good approach - can we do the same? Katja Klein Serial Powering vs. DC-DC Conversion 28