System Test Measurements
with DC-DC Converters
Katja Klein
1. Physikalisches Institut B
RWTH Aachen University
Tracker Upgrade Power Task Force
November 24th, 2008
Caveat
• Results are obtained with the APV25.
Other front-end chips might behave differently.
• Not all effects are fully understood.
• This is work in progress.
Katja Klein
System Test Measurements with DC-DC Converters
2
The APV25
f = 1/(250nsec)
= 3.2MHz
Katja Klein
System Test Measurements with DC-DC Converters
3
The APV25
1.25V
2.5V
*
* is connected to 2.5V since about 2000
Katja Klein
System Test Measurements with DC-DC Converters
4
The APV25
1.25V
2.5V
Katja Klein
System Test Measurements with DC-DC Converters
5
On-Chip Common Mode Subtraction
• 128 APV inverter stages powered from 2.5V via common resistor (historical reasons)
 mean common mode (CM) of all 128 channels is effectively subtracted on-chip
• Works fine for regular channels which see mean CM
• CM appears on open channels which see less CM than regular channels
• CM imperfectly subtracted for channels with increased noise, i.e. edge channels
pre-amplifier
V250
inverter
V250
R (external)
V125
strip
vIN+vCM
vCM
vOUT = -vIN
VSS
Node is common to
all 128 inverters in chip
Katja Klein
System Test Measurements with DC-DC Converters
6
System Test Setup
6.4
Katja Klein
6.3
6.2
• TEC petal with InterConnect Board
• Four ring-6 modules powered & read out
• Petal housed in grounded metall box
• Optical readout
• Optical control communication
• Thermally stabilized at +15°C
• Final components (spares)
• Official DAQ software
6.1
System Test Measurements with DC-DC Converters
7
Definitions and Analysis
• Raw (or total) noise = RMS of fluctuation around pedestal value
• Common mode (CM) = common fluctuation of subset of strips, calculated per APV
• The raw noise includes the common mode contribution
• One ring-6 module has four APVs a 128 strips = 512 strips
• Most tests performed in peak mode
• Modules are biased with 250V
{
A typical noise distribution
1 APV
Higher noise
on edge channels
Open channels
Katja Klein
System Test Measurements with DC-DC Converters
(e.g. from bias ring)
8
System Test with DC-DC Converters
Measurements with
commercial buck converters
with internal ferrite-core inductor
Katja Klein
System Test Measurements with DC-DC Converters
9
Commercial Buck Converter EN5312QI
Enpirion EN5312QI:
 Small footprint: 5mm x 4mm x 1.1mm
 High switching frequency: fs  4 MHz
 Unfortunately modest conversion ratio (r): Vin = 2.4V – 5.5V (rec.) / 7.0V (max.)
 Most measurements done with Vin = 5.5V  r(2.5V) = 0.45; r(1.25V) = 0.23
 Sufficient current: Iout = 1A
 Integrated planar inductor
Internal inductor in MEMS technology
Katja Klein
System Test Measurements with DC-DC Converters
10
Integration into TEC System
• 4-layer adapter PCB (W. Karpinski)
• Plugged between Tracker End Cap (TEC)
motherboard and FE-hybrid
• 2 converters provide 1.25V and 2.5V for FE-hybrid
• Input and output filter capacitors on-board
 standard caps; low ESR caps tested later, no improvement
• Input power external or via TEC motherboard
Katja Klein
System Test Measurements with DC-DC Converters
11
Integration into TEC System
L-type:
Larger flat PCB,1 piece
TEC motherboard
(InterConnect Board, ICB)
Front-end hybrid
S-type:
Smaller inclined PCB, 2 pieces
Katja Klein
System Test Measurements with DC-DC Converters
12
Voltage Ripple (L Type)
With load (I = 0.7A):
Without load:
Vin = 6V
Vin = 6V
High frequency
ringing
50mV
V2.50
V2.50
10mV
V1.25
V1.25
10mV
Ripple with
switching
frequency
200ns
 100mVpp high frequency ringing on input
 10mVpp ripple with switching frequency on output
Katja Klein
System Test Measurements with DC-DC Converters
13
Frequency Spectrum
Standardized EMC set-up to measure
Differential & Common Mode noise
spectra (similar to set-up at CERN)
PS
Load
Line impedance
stabilization
network
Load
Spectrum
analyzer
Spectrum
Analyzer
Copper ground
plane
Enpirion
2.5V at load
Common mode
Current probe
Enpirion
2.5V at load
Differential Mode
fs
Katja Klein
System Test Measurements with DC-DC Converters
14
Raw Noise with DC-DC Converter
 No converter
 No converter
Pos. 6.4
 Type L
 Type S
Pos. 6.4
 Type L
 Type S
 No converter
Pos. 6.4
• Raw noise increases by up to 10%
• Large impact of PCB design and
connectorization
• Broader common mode distribution
• Most optimization studies performed with L
Katja Klein
System Test Measurements with DC-DC Converters
 Type L
 Type S
15
Raw Noise on Different Positions
 No converter
Pos. 6.2
 No converter
 No converter
 Type L
 Type L
Pos. 6.4
Katja Klein
System Test Measurements with DC-DC Converters
 Type S
Preliminary
 Type S
Preliminary
Pos. 6.3
 Type S
Preliminary
On position 6.1 type L does not fit
 Type L
16
PCB Layout & Connectorization
---- No converter
Pos. 6.4
---- L type
---- S type
 Some variation within types, but
L and S are clearly different
---- No converter
Mean = 1.92548
 = 0.0314997
Mean = 2.08891
 = 0.0340944
---- L type with S-connector
---- Worst S type
Pos. 6.4 ---- Best S type
 Indication that difference comes from
impedance of “bridge“ connector
Katja Klein
System Test Measurements with DC-DC Converters
17
Edge Channels with DC-DC Converters
• On-chip CM subtraction: “real“ CM, i.e. before subtraction, visible on open and edge channels
• Huge increase of noise at module edges and open strips with DC-DC converter
• This indicates the real CM
 No converter
 Type L
Pos. 6.4
Katja Klein
 Type S
System Test Measurements with DC-DC Converters
18
Module Edge Strips
APV25 pre-amplifier
V250
V125
strip
bias ring
VSS=GND
[Mark Raymond]
connect to
V125
[Hybrid]
Katja Klein
• Edge strips are capacitively coupled to bias ring
• Bias ring is AC coupled to ground
• If V125 is noisy, pre-amp reference voltage
fluctuates against input
• Bias ring AC-coupled to V125  strip and
pre-amp reference fluctuate together, lower noise
System Test Measurements with DC-DC Converters
19
Comparison between Supply Voltages
 No converter
 L10
 L11
zoom on edge channels
 L10, 1.25V from converter
 L11, 2.5V from converter
• Either 2.5V or 1.25V line powered via converter; other line powered normally
• Overall noise increases when converter powers 2.5V
 1.25V powers ~only pre-amp; noise on 1.25V subtracted by CM subtraction
 2.5V used also after inverter stage; this contribution cannot be subtracted
• Edge strip noise increases more if converter powers 1.25V
Katja Klein
System Test Measurements with DC-DC Converters
20
Cross-Talk
Adjacent modules powered with EN5312QI converters
 study correlations between pairs of strips i, j (r = raw data):
corrij = (<rirj> - <ri><rj>) / (ij)
Pos .6.3
Pos .6.2
Pos .6.1
Pos. 6.4
Pos .6.3
Pos .6.2
Pos. 6.1
Strip number
Correlation coefficient
Pos. 6.4
Strip number
With converters on all positions
Correlation coefficient
Strip number
Without converters
Strip number
 High correlations only within single modules (= common mode)
 No cross-talk between neighbouring modules
Katja Klein
System Test Measurements with DC-DC Converters
21
Noise vs. Conversion Ratio (Vout/Vin)
• Vout fixed to 1.25V & 2.5V  change of Vin leads to change of conv. ratio r = Vout / Vin
• Output ripple Vout depends on duty cycle D (D = r for buck converter):
 Vout 
1
1 Vout
 Vout
1
1
(
V
in  Vout ) 

1

D


LCout fs 2 Vin
Vout
LCout fs 2
Mean noise per module
Pos. 6.4
Type L
Typical Vin
for system test
 Mean noise increases for decreasing conversion ratio
Katja Klein
System Test Measurements with DC-DC Converters
22
Combination with Low DropOut Regulator
• Low DropOut Regulator (LDO) connected to output of
EN5312QI DC-DC converter
• Linear technology VLDO regulator LTC3026
• LDO reduces voltage ripple and thus noise significantly
 Noise is mainly conductive and differential mode
• Would require a rad.-hard LDO with very low dropout
 No converter
Katja Klein
 Type L without LDO
 No converter
 Type L with LDO,
dropout = 50mV
 Type L without LDO
System Test Measurements with DC-DC Converters
 Type L with LDO,
dropout = 50mV
23
Other Commercial Converters: MIC3385
Are we looking at a particularly noisy commercial device?  Micrel MIC3385
- fs = 8 MHz
- Vin = 2.7–5.5 V
- footprint 3 x 3.5 mm2
 No converter
 Enpirion L-type
 Micrel L-type
 No evidence that EN5312QI is exceptionally noisy
Katja Klein
System Test Measurements with DC-DC Converters
24
System Test with DC-DC Converters
Measurements with
commercial buck converters
with external air-core inductor
Katja Klein
System Test Measurements with DC-DC Converters
25
External Air-Core Inductor
Enpirion EN5382D (similar to EN5312QI)
operated with external inductor:
From data sheet
• Air-core inductor Coilcraft 132-20SMJLB;
L = 538nH
• Ferrite-core inductor Murata LQH32CN1R0M23;
L = 1H
Air-core inductor
Ferrite-core inductor
6.60mm
10.55mm
5.97mm
Katja Klein
System Test Measurements with DC-DC Converters
26
External Air-Core Inductor
 No converter
Pos. 6.2
radiative part
 Internal inductor
 External ferrite inductor
 External air-core inductor
 Ext. air-core inductor + LDO
conductive part
 No converter
Pos. 6.3
(Conv. is on 6.2)
 Internal ind.
 Ext. ferrite ind.
 Ext. air-core ind.
• Huge wing-shaped noise induced by
air-core inductor, even on neighbour-module
• Conductive part increases as well
• Both shapes not yet fully understood
• LDO leads only to marginal improvement
Katja Klein
System Test Measurements with DC-DC Converters
27
Correlations
corrij = (<rirj> - <ri><rj>) / (ij)
APV 4
Pos. 6.4 (with conv.)
Per APV:
 two halfs of 64 strips
 strips within each half are
highly correlated (90%)
 but two halfs are strongly
anti-correlated (-80%)
 linear, see-saw like CM
 50% correlations also on
module 6.3 (no converter)
Pos. 6.3 (no conv.)
Katja Klein
System Test Measurements with DC-DC Converters
28
Radiative Noise from Air-Core Coil
• Module “irradiated“ with detached converter board
Mean noise
of 6.4
[ADC counts]
Reference
without
converter
• Increase of mean noise even if converter is not plugged  noise is radiated
• Noise pick-up not in sensor but close to APVs
Katja Klein
System Test Measurements with DC-DC Converters
29
External Air-Core Toroids
Toroid wound from copper wire
 No converter
 Solenoid
 Wire toroid
 Strip toroid
Pos. 6.4
Preliminary
Toroid wound from copper strip
• As expected, toroid coils radiate less than solenoids
• Significant improvement already observed with simple self-made toroid coil
Katja Klein
System Test Measurements with DC-DC Converters
30
Shielding the DC-DC Converter
• Enpirion with air-core solenoid wrapped in copper or aluminium foil
• Skin depth at 4MHz: Cu = 33m; Al = 42m
 No further improvement for > 30m, thinner foils should be tried
• No diff. betw. floating / grounded shield  magnetic coupling
• Shielding increases the material budget
 Contribution of 3x3x3cm3 box of 30m Al for one TEC: 1.5kg
(= 2 per mille of a TEC)




Katja Klein
No converter
No shielding
30 m aluminium
35 m copper
System Test Measurements with DC-DC Converters
 No converter
 Copper shield
 Aluminium shield
31
Distance to FE-Hybrid
 No converter
• The distance between converter and
FE-hybrid has been varied using a cable
between PCB and connector
• Sensitivity to distance is very high
• Conductive noise is decreased as well
due to filtering in the connector/cable
• Place converter at substructure edge?
 Type L
 Type S‘
 Type S‘ + 1cm
 Type S‘ + 4cm
Pos. 6.4
Type L with Solenoid
Katja Klein
Type S‘ with Solenoid
Type S‘ 4cm further away
System Test Measurements with DC-DC Converters
32
Combination of Measures
zoom on edge channels
Can get as good as without converter when toroidal coils are shielded
and combined with an LDO.
Katja Klein
System Test Measurements with DC-DC Converters
33
Summary and Worries
• With DC-DC converters powering the APV25, noise is an issue:
 noise ripple on power lines (conductive)
 rapidly varying magnetic near field (radiative)
• Possible countermeasures:
 Filtering: low pass, LDO, CM filter
 Shielding
 Distance
• A clever combination of these countermeasures will probably work
• These countermeasures add complexity, cost, mass to the system
• Noise-tolerant design required for FE-chip, hybrids, PCBs;
DC-DC conversion does NOT factorize from remaining system
• Noise performance should be robust and should scale with system size
Katja Klein
System Test Measurements with DC-DC Converters
34
Back-up Slides
Katja Klein
System Test Measurements with DC-DC Converters
35
System Test with DC-DC Converters
Measurements with
custom DC-DC converters
Katja Klein
System Test Measurements with DC-DC Converters
36
The CERN SWREG2 Buck Converter
• Single-phase buck PWM Controller with Int. MOSFET dev. by CERN (F. Faccio et al.)
• HV compatible AMI Semiconductor I3T80 technology based on 0.35 m CMOS
• Many external components for flexibility
• PCB designed by RWTH Aachen University
Vin = 3.3 – 20V
Vout = 1.5 – 3.0V
Iout = 1 – 2A
fs = 250kHz – 3MHz
S. Michelis (CERN) et al., Inductor based switching DC-DC converter for low voltage power distribution
in SLHC, TWEPP 2007; and poster 24 at TWEPP 08.
Katja Klein
System Test Measurements with DC-DC Converters
37
The CERN SWREG2 Buck Converter
• PCB located far away from module
 only conductive noise measured
• SWREG2 provides only 2.5V to FE-hybrid
• Noise increases by about 20%
• 8-strip ripple structure understood to be
artefact of strip order during multiplexing;
converter affects readout stages of APV
Vin = 5.5V
Katja Klein
No converter
2.5V via S type
0.60 MHz
0.75 MHz
1.00 MHz
1.25 MHz
System Test Measurements with DC-DC Converters
Order in APV
Physical
afterorder
MUX
38
The LBNL Charge Pump
• Simple capacitor-based step-down converter: capacitors charged in series and
discharged in parallel → Iout = nIin, with n = number of parallel capacitors
• At LBNL (M. Garcia-Sciveres et al.) a n = 4 prototype IC in 0.35 m CMOS process
(H35) with external 1F flying capacitors has been developed (0.5A, 0.5MHz)
P. Denes, R. Ely and M. Garcia-Sciveres (LBNL), A Capacitor Charge Pump DC-DC Converter for
Physics Instrumentation, submitted to IEEE Transactions on Nuclear Science, 2008.
Katja Klein
System Test Measurements with DC-DC Converters
39
The LBNL Charge Pump
• Two converters connected in parallel: tandem-converter
 fs = 0.5MHz (per converter)
 In-phase and alternating-phase versions
• Tandem-converter connected either to 2.5V or 1.25V
• Noise increases by about 20% for alternating phase
 No converter
 In-phase on 2.5V
 Alt-phase on 2.5V
Katja Klein
System Test Measurements with DC-DC Converters
40
The Buck Converter (Inductor-Based)
The “buck converter“ is the simplest inductor-based step-down converter:
Switching frequency fs:
fs = 1 / Ts
Duty cycle D:
D = T1 / Ts
Convertion ratio r < 1:
r = Vout / Vin = D
• Can provide relatively large currents (several Amps)
• Technology must work in a 4T magnetic field
 since ferrites saturate at lower fields, air-core inductors must be used
• Need to find compromise between high efficiency (low fs) and low noise (high fs)
• Discrete components (coil, capacitors, ...) need space and contribute to material
Katja Klein
System Test Measurements with DC-DC Converters
41
The Charge Pump (Capacitor-Based)
Step-down layout: capacitors charged in series & discharged in parallel
n = number of parallel capacitors
Iout = nIin
r=1/n
•In
→
• Inductorless technology
• Everything but capacitors is integrated on-chip  low mass, low space
• Provides relatively low currents (~ 1A)
• Fixed “quantized“ conversion ratio (e.g. r = ½)
• Output voltage cannot be controlled by feedback loop
• Switching noise
Katja Klein
System Test Measurements with DC-DC Converters
42