1. Motivation & History for HV Mux Radiation Hardness & Irradiations
Radiation Hard GaNFET High Voltage Multiplexing (HV Mux) for the ATLAS Upgrade Silicon Strip Tracker
Motivation & History for HV Mux
Radiation Hardness & Irradiations
HV Mux Circuits
Other Rad-hard Components
Reliability
Summary
D. Lynn (BNL), Bart Hommels (Cambridge), Giulio Villani (RAL), Miguel Ullan (CNM) for the ATLAS Collaboration.
Topical Workshop on Electronics for Particle Physics, UCSC, September 14th, 2017

2. Motivation
Each module in present ATLAS SCT (the silicon strip detector) has its own independent HV power supply.
Cable space limitations do not permit each module in the stave to have its own HV power supply (more than 10k modules in Atlas Strip upgrade barrel vs ~ 2k in present barrel).
Solution: Have multiple modules connected to HV bus with slow controlled switch that can disconnect malfunctioning sensors.
Two variations of 14 modules on a stave with HV switches to disconnect malfunctioning sensors
Slow controlled switch

3. Atlas Strip Modules and Staves
Atlas Strip Module with HV Mux
A barrel module consists of a 10 cm x 10 cm silicon sensor with one or two hybrid circuits containing the ABC130 readout ASICS.
Module has a powerboard that contains DC-DC converter circuitry for LV power and the HV Mux circuit that can disconnect the silicon sensor from the HV power bus.
A stave is a collection of modules on a carbon composite support and cooling structure.
Thermo-Mechanical Stave Prototype

4. History of HV Mux Project
Started looking for HV switches almost 7 years ago (2010).
Requirements were that they can switch > 500 V, operate in 2T magnetic field, and survive ~ 1 x 1015 neq /cm2 and ~ 30 Mrad (or perhaps 1.6 x 1015 neq /cm2 and ~ 50 Mrad).
Early results with EPC (Efficient Power Conversion) Gallium Nitride transistors (GaNFETs) were very promising V SiC devices looked promising as well.
Organized program to irradiate a variety of GaN, SiC transistors and Si devices (this has been the main focus of the HV Mux project).

5. Present Status
In the last couple of years we have converged on GaNFETs from Efficient Power Conversion (EPC), GaNSystems, and Panasonic as only GaN technology survived high doses.
EPC. Their highest voltage device is now only 200V. A 450V device was available and we tested it and had plans for a cascade circuit to double the voltage. But the 450V device was taken off the market.
GaNSystems. They have a 650V device that survived our irradiations very well. This is an extremely strong candidate.
Lack or resources required we pursue irradiation campaign of only a single candidate. We chose the Panasonic (600 V) GaNFET.

6. GaNFET Operation and Radiation Hardness
GaN is high bandgap (3.4 eV) semiconductor suitable for higher voltage applications.
The devices behave similar to a MOSFET, but have no gate oxide.
Unlike a MOSFET, the inversion channel is created by a piezoelectric effect created by the lattice mismatch between the i-AlGaN and i-GaN layer.
Drawings from : “Current-collapse-free Operations up to 850 V by GaN-GIT utilizing Hole Injection from Drain”, S. Kaneko et al; Proceedings of the 27th International Symposium on Power Semiconductor Devices & IC's, May 10-14, 2015, Kowloon Shangri-La, Hong Kong

7. GaNFET Operation and Radiation Hardness
Possible reasons for radiation hardness:
Our irradiations have shown GaNFETs to be very radiation hard (although some failures have occurred, mostly in the off-state with Vgs = 0 V and high voltage on the drain).
Bulk damage due to n (and p+/p) does very little to the GaN devices, because the carrier concentration/mobility in the inversion channel is high enough
Total Ionizing Dose (TID) does very little because of the small thickness of the i-AlGaN, that effectively makes the generated charge in it 'tunnel away' (like in a thin GOX standard Si MOS device).

8. EPC 200 V Irradiations
EPC1012 was the first generation 200 V device, EPC2012 was the second generation 200 V device.
At Los Alamos we typically irradiate devices either in the on-state or off-state.
On-state mean Vgs =2.5 V (or 4 or 5 V), Vds = 0 V. Off-state means Vgs = 0, and Vds = 175 V.
All failures believed to be by ESD/handling.

9. EPC 450 V, GaNSystems 650V Irradiations
On-state mean Vgs =2.5 V, Vds = 0 V. Off-state means Vgs = 0, and Vds = 400 V or 500 V.
All devices performed well. 450V device was pulled of the market.
GaNSystems a promising candidate.

10. Panasonic 600V Irradiations
On-state mean Vgs =2.5 V, Vds = 0 V. Off-state means Vgs = 0, and Vds = 500 V.
Strong candidate despite two failures at the very high dose in the off-state at Los Alamos.

11. Comments on GaNFET Irradiation Results
Over three GaNFET vendors and many devices, we has seen that the devices appear extremely radiation hard to total dose, and in particular when powered in the on-state (the normal operational mode for the detector).
There have been some failures in devices irradiated in the off-state (at extremely high dose levels). On-state failures have not been seen.
Plan over the last year has been to try to irradiate more parts of the same device (Panasonic) in order to accumulate statistics.
We obtained 130 die for that purpose. The last irradiations will begun this month (September) and we expect results by the end of this year.

12. Panasonic 600V Bare Die Irradiations
Began campaign in 2016 to irradiate > 100 die at a variety of facilities
Sept-Oct 2017
Sept 2017
Note impressive results at Birmingham. We have decided to focus on on-state irradiations as we realized the reliability of the devices matters much more in the on-state (off-state implies the sensor has already failed.).

13. Why Choose Panasonic over GaNSystems
Some earlier irradiations at Karlsruhe (25 MeV protons) with GaNSystems had failures. However, we now believe this was due to mishandling.
Some GaNSystems devices we purchased seemed to have failed when mounted on circuit boards and powered up. Now we know there were soldering errors.
I misread the very impressive Los Alamos results and thought they had failed. I didn’t realize my error until months later.
We are not able to get GaNSystems parts as die, although it is possible (based upon my conversation with GaNSystems) to get them in a smaller custom package if we worked with their packaging company.
Panasonic devices seemed to have a little more consistency in their characteristics, such as threshold voltage.
Conclusion: Errors on our part led us to believe GaNSystems was less radiation hard. We had already chosen Panasonic which has been shown to be very radiation hard at our expected doses.

14. HV Mux Circuit LBNL V2 PowerBoard with HV Mux RC Filters Detector
Schottky Diode Multiplier for Gate Bias
50 kHz Oscillator
GaNFET
Gate Overvoltage Protection Zener

15. HV Mux Circuit Switching On and Off Vbias
Simple LTSpice Simulation

16. Additional Radiation Hard Components
Schottky diode array (4 diodes) BAT54SDW from Diodes, Inc and Micro Commercial survived 6 x n/cm2 with little change. But we are moving to Diodes, Inc BAS70DW Schottky diode array (lower parallel capacitance).
Bourns CD0603-Z5V1 Zener survived 6 x n/cm2 with little change. But we are moving to Rohm Zener VDZ4.7B as the Bourns Zener is not recommended for new designs.
So we believe we have the radiation hard components for the 500V circuit. We will still continue with irradiating their replacements.

17. HV Mux Reliability
We are still in the process of trying to understand overall reliability of HV Mux.
Focus has been on irradiation campaigns. We will continue these campaigns.
For production, the plan is to buy a single sequential batch of 20,000 die (in the form of diced wafers). We will extensively irradiate ~ 5% (1000 die).
The idea of buying a single batch is so we know there are no process changes between wafers.
We are also in conversations with Panasonic to understand the results of their own internal reliability data and conclusions.
We will also try to understand the reliability of the overall circuit. To first order, the reliability of the circuit is the product of the reliability of:
Oscillator circuit (which is part of a custom ASIC from University of Penn called the AMAC)
The multiplier circuit which is Schottky diodes and capacitors.
Panasonic GaNFET

18. HV Mux Summary
For the ATLAS Phase II silicon strip upgrade: We began looking into a solution around 2010 to find a way to disable malfunctioning (i.e. high current) sensors from a HV bus in order to allow continued operation of the remaining sensors on the bus.
The devices need to survive high radiation and a 2T tesla field. This limits technological choices.
We converged on a Panasonic Gallium Nitride Transistor (GaNFET) as our primary switch candidate.
We’ve established that Gallium Nitride switches tend to be radiation hard.
We developed control circuits for the GaNFET that are also radiation hard.
We’ve begun to understand overall reliability of HV Mux but have more work to do.

19. Additional Slides
Other Candidate Switches Once Considered
700 V (or 1000 V Variant)
Radiation Results for Blocking Diode
Vbias 700 V Panasonic Radiation Hard Cascade Circuit
Results of 6 x 1015 n/cm2 Irradiation at Ljubljana in 2015
Long Term Cold Testing of Devices
Oscilloscope Measurements on Prototype HV Mux Circuit
Failure Probability Analysis
Panasonic Leakage Current Post Irradiation

20. Other Candidate Switches Once Considered
Radant MEMS switches. Require high voltage on the gate. Too low Vds rating. Custom development too expensive.
SiC JFET Semisouth V rated. Showed some potential at ½ x 1015 p+/cm2. But company went out of business.
Not seriously considered
Relay. Essentially what we want, but won’t work in magnetic field.
Mosfet. Power Mosfets have too thick of an oxide and therefore are not radiation hard. NMOS FETs in 130nm technology, for example, in principle should work. Need to stack 500 V/2.5 V = 200 transistors. Would need multiple substrates. Control issues (But still maybe worth a look?).
BJT. Not sufficiently radhard at these levels.

21. 700 V (or 1000 V Variant)
Possibly the HV requirement for the ATLAS strips may increase to 700 V.
This circuit stacks two GaNFETs for higher voltage
Need a blocking diode that can withstand > Vbias/2. Have not yet found one in a radiation hard technology
Rad-hard version of a 700 V circuit is in backup slides; but this version requires more space .
Still need to find 400 V (600 V) rad-hard blocking diode

22. Radiation Results for Blocking Diode
Microsemi 82 V Zener (Vds protection in cascade circuit) failed 6 x n/cm2. Use RC cicuits on drain source to balance voltage sharing between GaNFETs in 700V circuit.
Sensitron 1C V diode (blocking diode) has too large increase in forward resistance at 6 x n/cm2 . But still a candidate. Waiting for 2 x n/cm2 results.
NXP BAS V blocking diode shows large increase in forward resistance. Fails.

23. Vbias 700V Panasonic Radiation Hard Cascade Circuit
Blocking Diode Replacement
C1 47 nF, 630 V, 1206
C3 15 nF, 630 V, 1206
C11 15nF, 630V, 1206
C14 15 nF, 630V, 1206
C4, 15 nF, 1000V, 1206
C9, 15 nF, 1000V, 1206
Set C1 ≈ C3 || C14 || C11
HV Zener Replacements
C11
C14 C1
Question is whether we have room for this version. Capacitors in labeled in yellow take up a lot of space. C3 C4 C9

24. Results of 6 x 1015 n/cm2 Irradiation at Ljubljana in 2015
HV diode BAS521. Semi-operational at 6 x 1015 n/cm2 fluence.
EPC Works at 6 x 1015 n/cm2 fluence.
Schottky diodes. Works at 6 x 1015 n/cm2 fluence.
Zener diode 5V1. Works at 6 x 1015 n/cm2 fluence.
GaNSystems GS66502B 650 V, Normally-Off. Works at 6 x n/cm2 fluence.
Panasonic PGA26E19BV, 600 V, Normally-Off. Works at 6 x 1015 n/cm2 fluence.

25. Long Term Cold Testing of Devices
Used 10 Panasonic GaNFETs previously irradiated with pions at PSI, Switzerland.
Run in off-state ( Vgs = 0 V, Vds = 500 V). Measure leakage current Ids .
Test conditions: less than 30% RH at -20°C.
Ran each device for 2000 hours.
Increased Vds to 600 V. Have run for 400 more hours each.
Leakage current remains low (~ 4-8 nA).
Plan is to keep devices running next few years.
Will shortly had additional devices running in the on-state ( Vgs = 2.5 V, Vds = 0 V).

26. Oscilloscope Measurements on Prototype HV Mux Circuit
Vbias
Vbias
50 kHz Oscillator
ON transient
OFF transient
Time plot of output voltage (i.e. measured at HVout node).
HVin set to – 550V
Turning ON Settling time  45 ms MAX, including turning ON of OSC
Turning OFF Settling time  50 ms MAX including turning OFF of OSC
Maximum Leakage current through HV switch in OFF state:  V

27. Failure Probability Analysis; Should We Use HV Mux?
PHVO (or simply Hvo) is probability high voltage circuit fails open (plot Hvo = 1%, 2%, 5%)
Plot Mean Percentage of Failed Modules as function of (unknown) probability a sensor goes into breakdown or shorts with and without the use of HV Mux.
Plot for N = 4 modules on single HV line (default) and N = 7 modules (previous baseline)
HV Mux can be viewed as insurance against > 2% of sensors failing

28. Panasonic Leakage Current Post Irradiation
Devices 1 and 2 were irradiated to 5.8 x 1015 p+ /cm2 in the on-state.
Leakage currents remains very low.
Not uncommon with the various GaNFETs for gate leakage to exceed source leakage