Transcript Folie 1

Integration Aspects
of DC-DC Converters
Katja Klein
1. Physikalisches Institut B
RWTH Aachen University
TUPO, June 10th, 2009
Outline
• Conversion ratio & output current
• Dimensions and weight of buck converters
• Material budget
• Cooling requirements
• Shielding requirements
• Should the DC-DC converter be part of the module?
• Possibility of integration for various module proposals
• Provision of two operation voltages for CBC
• Discussion of four options
• Conclusion & recommendation
Assumption: GBT powered from outside of sensitive volume
Katja Klein
2
Comparison of Layouts
Layout
FEPower
Link-Power
Total Power
# of
Modules
FE-Power
per module
Long barrel double
stack (Marcello)
100kW
25kW $
125kW
20 000
4 - 9W
Hybrid
layout §
(Duccio)
tracking
12.5kW 2.9kW &
43kW
10 040
0.94 - 1.9W
trigger
12kW
1 568 *
1.3 – 9W #
Cluster
width
Fabrizio;
barrel only
20.9kW 2.3 – 14.1kW ° 23.2 – 35.0kW
14 037
1.25W
Duccio;
full outer tr.
56.5kW ~ 20kW
13 008
1.1 – 9.4W
15.6kW &
~ 75kW
All power numbers include a DC-DC efficiency of 80%
§ Variant with 2 long barrel pT layers and tracking-only endcaps
$ assuming 10Gb/s GBT-like link, 2W per link
& with 2W/GBT
° depends on optical module (GBT vs. MZM), larger number for GBT (3W per GBT)
* for A = 85cm2
# depends strongly on module proposal
Katja Klein
3
Total Power Consumption
• Total power consumption limited by heating up of water-cooled cable channels
• Today the total current in cable channels is 15kA
• Upper limit would have to be determined by measurements on mock-ups of
hot spots in cable channel (Hans Postema)
• 10-20% more might be possible, but probably not more? (Hans Postema)
• Can calculate maximum power consumption for certain convertion ratio r = Iin / Iout:
E.g. for r = 1/10 and 80% efficiency: Pmax = 150kA x 1.2V x 0.8 = 144kW
• Can estimate the necessary conversion ratio for a given power consumption:
r = 15kA / Iout
P = Uout x Iout (includes already converter efficiency of 80%)
r = 15kA x Uout /P
Katja Klein
Layout
Total Power
Operating voltage
Conversion Ratio
Long barrel
double stack
125kW
0.9V
1/10
Hybrid
strawman
43kW
1.2V
0.4
4
Conversion Ratio from Cable Specs
• Assume only 1 000 LICs can be used to power the modules (reason: links)
• Spec of LICs: Umax = 30V, Imax = 20A (return)
• Calculate mean number of modules per LIC
• Calculate mean current per LIC
• Estimate necessary conversion ratio
• In reality, could try to level out (but then granularity becomes an issue)
Layout
# of
Modules
Power per
module
# Modules Current per LIC Conv. ratio
per LIC
(worst case)
Long barrel double
stack
20 000
4 - 9W
20
200A
1/10
Hybrid
tracking
strawman
trigger
10 040
0.9 - 1.9W
12
19A
1
1 568
up to 9W
12
120A
1/6
Katja Klein
5
First Conclusion
• Buck converters are needed at least for trigger layers
• Charge pumps are no option for some approaches (max. current ~ 1A)
 Currents to be provided too big for a single charge pump per module
 Charge pump per chip not feasible (90nm, no space for capacitors, ...)
• Discuss in the following the integration of buck converters
• Come back to charge pumps later
Katja Klein
6
Dimensions and Weight of Buck PCBs
Smallest Aachen PCB (V1):
Area: 2.3cm2
Height: 10mm
Weight: 1.0g
12mm
19mm
Aachen PCB with
lowest noise (V3):
12mm
Area: 3.2cm2
Height: 10mm
Weight: 1.1g
Numbers are without connectors
Katja Klein
27mm
7
Dimensions and Weight of Buck PCBs
CERN PCB (proposal):
INDUCTOR
SMD
ASIC
SMD
SMD
SMD
1.5-2 cm
1.5-2 cm
 Area (currently) needed per buck converter PCB: 2-4cm2
• Some further minimization probably possible (e.g. remove connectors)
• But filter capacitors are necessary
• Coil must have a certain inductance ( noise) and low DC resistance ( efficiency)
Katja Klein
8
Material Budget
• TEC, conversion ratio 1/8, eff. = 80%, current power consumption, 1.2V only (Aachen)
• One buck converter per module, located close to module
Total MB of:
TEC modules
TEC Converters
TEC electronics & cables: - 29%
Original TEC
TEC with
buck converters
r = 1/8
 With above assumptions, buck converter close to module saves material
(caveat: savings are half due to DC-DC conversion, half due to methodology)
Katja Klein
9
Material Budget
Buck converters with r = 1/8,
located at petal rim:
TEC electronics & cables: - 18%
Buck converters with r = ¼ at petal rim,
one charge pump with r = ½ per chip
TEC electronics & cables: - 24%
 Buck converter close to module gives largest saving for TEC
 Desirable to repeat study for barrel geometry and two operation voltages
Katja Klein
10
Cooling Requirements
• Converter efficiency ~ 80%
• Heat to be dissipated ranges from 150mW (outer tracker module with 1 hybrid)
to ~ 2W (3D-integrated stacked module, inner layers)
• A contact to the cooling system should be foreseen
Katja Klein
11
Shielding Requirements
Measurements with solenoid coil
(worst case)
• Measurements show that shielding the whole converter helps against EMI from coil
 Shielding only the coil was not so efficient (reason not completely understood)
• New Aachen boards need not be shielded anymore in our system test set-up
• Requirements depend strongly on distance to FE-electronics
(plus technical details of converter and coil...)
• Recommendation at this point: a DC-DC converter on the module should be shielded
 30m of Aluminium worked fine (no improvement with thicker shields)
 Details would have to be worked out and tested
Katja Klein
12
Integration of Buck Converter (I)
Arguments for buck converter on separate PCB, close to module:
• Very limited space on most proposed hybrids  size less critical
• Larger distance preferred for EMI anyway (also damping of ripple?)
• Converter development completely decoupled from hybrid and module development
 No common deadlines, can optimize converter design as needed (even late)
• Different hybrids for different module proposals  avoid involvement of many groups
• PCB could be developed, manifactured and tested standalone
• Easier for cooling? (module cooling is difficult enough without converters)
Arguments for buck converter on the module/hybrid:
• Less mass (avoid connectors & connection between converter and module)
• Power regulation closer to FE-ASICs (only relevant if no LDO)
• Could have pluggable PCB on hybrid, but then connectors are needed (mass)
• Noise effects can be tested more easily (don‘t need additional PCB)
Katja Klein
13
Outer Tracker Module Proposal
TCS I/O PLL
2 x 4-MUX + LCDS driver
each output 160Mbit/s
DC-DC
shielded micro-twisted pairs I/O
DC-DC out 2.5V
Sensor HV
• CBC-power ~ 0.75W per hybrid;
i.e. 0.75W or 1.5W per module
• Plus DCU, PLL, DC-DC inefficiency,
GBT-port, MUX, LCDS-driver
• No motherboards
• Upper part of hybrid ~ 2.5cm x 1cm,
no space for buck on this hybrid
• Some space between hybrids;
but routing of input & output?
• Integration of buck on rod level looks
more practical and elegant
Katja Klein
2.5cm
8x CBC 2x 128ch
wire bonded
40Mbit/s out each
DCU
Sensor with 4x2.5cm strips
2x 1024 @95um pitch
integrated pitch adaptor
14
Outer Tracker Module Proposal
• Indeed the buck converter must be able to provide several Amps
(as anyway needed by pT-modules)
• Could save material by combining two one-hybrid modules into one unit
• Could even consider to power two two-hybrid modules (3W) with one converter
• Loose two modules if converter fails
• No other drawbacks from power point of view
Katja Klein
15
Vertically Integrated Hybrid Module
• 130 or 90nm
• Communication through vias in
ROC and interposer (3D-integration)
• No motherboards
• FE-power 4-9W per stacked module
• Up to 10A per stacked module
• Need at least two buck converters
per stack (better more)
 module needs to be “partitioned“
• No space on module; no hybrid
• Modules integrated onto “beams“
• Buck converters must be integrated
into beam structure
• Shielded space already foreseen
• Discussions between Fermilab &
Aachen started
Katja Klein
16
Trigger Module (Sandro)
1 Modul:
• 90nm
• Sensor size = 4.8cm x 4.8cm
• Hybrid ~ 1cm x 4.8cm
• Power per pT-module = 2.6W
• I per modul ~ 3A
• No space for buck converter
(unless hybrid is considerably increased)
• Practical issues (fabrication?)
• Again, integration into support
structure seems more feasible
• How would support structure look like?
1 Chip
Katja Klein
17
Trigger Module (Geoff et al.)
data out
control in
26mm
80mm
• Sensor size ~ 2.6cm x 8.0cm
• Hybrid ~ 1cm x 4cm
• 130nm
• Power per pT-module ~ 1.3W (similar to outer tracker module)
• No space for buck converters, unless hybrid is considerably increased
Katja Klein
18
Integration of Buck Converter (II)
• There is a tendency to avoid motherboards at all
 Outer tracker module, vertically integrated double-stack proposal, others?
• This goes hand in hand with rather minimalistic hybrids of a few cm2
• All existing or planned buck converter PCBs need an area of 2 - 4cm2
• Suggestion: a separate buck converter PCB close to the module,
e.g. inside the beam (for double-stack approach) or on the rod/stave
 converter needs cooling contact – probably not too dificult then
 need short power cable between converter PCB and module
 Could/should be designed such that it fits with all proposals/applications:
 Version with 1.2V and 0.9V for CBC
 Version with two (or three) buck converters for very high-power trigger modules
 Version with 1.2V and 2.5V for GBT, for PP1 or bulkhead
Katja Klein
19
Provision of two Operating Voltages for CBC
• Vana = 1.2V, possibility to have Vdig < Vana (~ 0.9V)
• P = 64mW per Chip with 1.2V
(26mW analog power, digital power would be halved with U = 0.9V)
• Both analog and digital currents ~ 20-30mA per chip
• How to provide the two voltages? Options:
1. Use the two LV conductors in LICs and two separate buck converters

Same conversion ratio for both bucks

Power supplies must provide two voltages
2. Provide one common input voltage, use two separate buck converters

Different conversion ratios for bucks

Lower power losses than option 1.
3. Derive Vdig from Vana with linear regulator

Method with lowest efficiency
4. Derive Vdig from Vana with charge pump (ratio 4:3)

Option with lowest mass and space requirements

Brings us to more general question: do we want to use charge pumps, and how?
Katja Klein
20
Option 1: Only Buck Converters
• Conversion in one step
• Assume buck converter close to the modules with r = 1/6 or smaller (as needed)
• If the necessary conversion ratio can be realized in one step for all proposals
must be studied with new ASIC prototypes! (issue of switching losses)
• Do not use charge pumps  no additional chips on the FE-hybrid or inside CBC
• Must find space for 1 or 2 buck converters (as many as operating voltages)
either on your module/hybrid or (preferred!) on your support structure
• If decided to put buck on support structure, module design can proceed
completely independently
• Could fit with all proposals
• Maximal current per buck converter to be understood, of the order of 4A
 Looks tight for double-stack proposal, must find reasonable partitioning
Katja Klein
21
Option 2: Buck + Charge Pump per Module
• Could be necessary if conversion ratio cannot be provided in one step
 Buck converter with r  ¼ close to module; charge pump with r = ½ per module
• Is however NOT compatible with any pT-module (due to current requirements)
• (Only) Possible useful application: provision of Udig for CBC for one FE-hybrid
 conversion ratio 4:3, current ~ 300mA
 less material than two buck converters (but not half!)
 on cost of higher complexity
• Space for buck converter: see option 1
• In addition need space for 1 chip plus capacitors (details to be worked out)
 on FE-hybrid
 or even on buck PCB?
• Such a chip is currently not being developed
Katja Klein
22
Option 3: Buck + Charge Pump On-Chip
• Assume charge pump is integrated into read-out ASIC
• Concerns raised by Mark: substrate noise, constraints on layout, space for passives
• Could be necessary if conversion ratio cannot be provided in one step
 Buck converter with r  ¼ close to module; charge pump with r = ½ per module
• Seems NOT compatible with some pT-modules (technology, space for passives)
• Possible useful application: provision of Udig for CBC
 conversion ratio 4:3, current ~ 20mA
 less material than two buck converters (but not half!)
• Alternative: derive both voltages with charge pumps
 conversion ratio 1:2  one capacitor per voltage (100nF, 0201?)
 need LDO for analogue power
 can switch on/off single read-out ASICs
 how to power auxiliary ASICs (PLL, MUX, LCDS driver, ...)?
• Space for buck converter: see option 1
• In addition need space for capacitor(s) close to CBC on FE-hybrid
• Design block for 60mA in 130nm being developed by CERN/Atlas
Katja Klein
23
Option 4: Buck + Sep. Charge Pump per Chip
• Assume now separate charge pump chip per readout-ASIC
• No substrate noise, no constraints on CBC layout
• Space for passives still needed, plus space for charge pump chips
• Very small chips to be integrated onto hybrid – possible but cumbersome?
• Looks NOT compatible with some pT-modules (technology, space)
• Possible useful application: provision of Udig for CBC
 conversion ratio 4:3, current ~ 20mA
 less material than two buck converters, but more than option 3
• Alternative: derive both voltages with charge pumps
 conversion ratio 1:2  one capacitor per voltage (100nF, 0201?)
 need LDO for analogue power
 can switch on/off single read-out ASICs
• Space for buck converter: see option 1
• In addition need space for 1 or 2 chips plus capacitor(s) close to CBC
Katja Klein
24
Conclusion & Recommendation (my opinion)
• Buck converters cannot be avoided (but charge pumps can)
• No motherboards and no or very small hybrids
 integrate buck converter onto separate small PCB
• Cannot decide today if charge pumps are needed, keep option open
• Abandon option 4 (separate charge pump chip per read-out ASIC)
• Prepare for option 1 with buck on support structure
• Explore and do not exclude options 2 & 3
 Allow some space for charge pump chip plus caps on hybrid
 Allow some space for caps close to CBC
 Integrate charge pump block offered by CERN group into CBC
 in a way that it can be bypassed
 would learn a lot about option 3
 this is an opportunity to make real progress!
Katja Klein
25