LBNL-2016: Instrumentation

From CMB-S4 wiki
Revision as of 13:58, 9 March 2016 by Nils.halverson (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigationJump to search

Notes from Wednesday session (taken by NWH):

Kent Irwin: Detector interfaces and questions

Antenna to room temperature readout interconnects: How many and where are they? Complexity cost associated with interconnects.


  1. 1-300 K architecture
  2. Mux element - volume, mass, power dissipation
  3. Pixel to mux interconnects, complexity -- do we need R & D on bump bonds, etc.

Need people to explore the following mux/detector options:

  1. TDM+
  2. FDM+
  3. MKID's
  4. RF TES's (microwave SQUIDs, parametric amps, JJ mux)

Need to quantify costs, power flow, complexity, power dissipation.

Adrian: Costs need to be broken down into:

  1. Startup cost and risk
  2. Production cost and risk
  3. Integration cost
  4. Testing cost

Kent: There is a cost per production pixel that is way to low, because it means you spent way too much in R&D to bring down cost.

We should figure out what we know.

Mike N.: Should we discuss timelines? Need to demonstrate key performance characteristics.

Jamie pointed out yesterday hat if there are particular technologies that we should push right away, we should write white paper and proceed to take advantage of this opportunity.

Collaboration process discussion:

Is there communication with DOE that needs to happen during prelim R&D? John C.: Timeline 2019, 2020 CD1, have to make all technology selections by then. Natalie R.: Some details on readout, etc, can be worked out between CD1 and CD2.

Decision: We need to target R&D done and make technology decision by 2019.

Need to give DOE a heads up on budget before then.

Adrian: Could identify and pursue high/med/low risk paths in parallel.

Action items:

  • Organize people and timeline for studies (Kent, Zeesh, Kam, Steve, Sarah, Toki, Darcy, Akito, Amber, Brad).

Deadline: will report to cosmic visions group and wiki by 4/16/16.

Charles: Costs should include requirements imposed on the rest of the system (e.g., magnetic shielding).

Whiteboard capture irwin 20160309.jpg

Clarence Chang: Multichroic pixels

We should look at drivers/costs for multichroic detectors.

Adrian: Massive multichroic could be disruptive and enabling. Put in a lot of R&D cost up front, but then big payoff.

Nils: But of course this approach also has risks -- need to ensure high yield to get the performance gains.

Jeff: We've already fielded dichroic arrays with 85% yield. Octave bandwidth is achievable.

Clarence, all: Can do study of MC number of channels optimization. But needs to take into account loss in mapping speed due to pixel size in f lamba (assuming no multi scale pixels), and telescope and readout cost.

Kam and Nils: Are we readout limited? What is advantage in that case of MC pixels?

Akito: Straightforward to calculate cost of inefficiencies

Adrian: Is there a fixed design that we field, or are we allowed to change arrays 3 years into observing? If so, we can learn about foregrounds and feedback into upgrade.

Jeff: If we are going to go for mixed array of large/small telescopes, will need different technologies for the two apertures.

Steve: Need to do modeling with realistic sensitivity, efficiency.

Adrian: Need to optimize resolving power R.

Kent: HWP will affect choice of bands/bandwidth.

Adrian: Broadband AR coatings are a major consideration.

Jeff: Know how to make 4:1 AR coating, but not 4:1 HWP.

Nils: Need to include risk in study since this will change optimization outcome

Steve: Risk too difficult to quantify and not needed now, but do need realistic efficiencies using existing measured performance.

Charles: Mapping speed is convenient, but overly idealistic. Limitations for B-mode problem are foregrounds and systematics. If you are not taking these into account, you are missing the boat. Suspect that frequency coverage will be important to decide, and difficult to change once observations have begun. Noise levels accounting for atmosphere are important to take into account. Suppose you had to go to 10 GHz for foregrounds. That would change study/instrument design.

John: What are practical limitations (loss, atm), and what research can we do to mitigate?


Brad: Trying to decide where we need R&D. We need to enumerate status of various techologies in lab and field, state NET's, optical efficiencies, etc.

Mike: We should write down systematics associated with all the existing detector technologies.

Adrian: Type of pol modulators affects requirements for detector systematics.

Mike: Let's keep systematics quantifications separate for detectors, pol modulator, telescope optics. Then can find end to end systematics for a variety of combinations.

Jeff: When we are calculating mapping speed, need to use practically realizable pixel sizes, atm windows, etc.

Keith: We should explore low frequency limit,

Action items:

  • Specify acheivable bandwidths for broadband optics (Zeesh (organizer),Jeff, Toki)
  • What resolving power is desirable/practical (Jeff (organizer), Toki, Akito, Kimmy)
  • Need to quantify systematics (Mike (organizer), Amy, Maria, Akito, Kimmy, Jeff, Nils)
  • Study mapping speed/frequencies/MC (Akito (organizer))

Deadline: will report to cosmic visions group and wiki by 4/16/16.

Brad Johnson: KIDS

MKIDs advantages:

  • high multiplexing factors, smaller number of processing steps, fast time constants, low power consumption, cost?
  • LEKID: change qp density by breaking cooper pairs, changes kinetic inductance. capacitively coupled

Noise sources: Dominated by photo noise above 5e-12 W incident power.


  • Commercial fabrication, turnaround 1 wk, cost ~$2k.
  • Result: measured noise Got 20-30 uK rt s with 4K load in lab.
  • Has coherent and broadband sources to illuminate detectors with collimated light.
  • < 0.1 Hz 1/f knee

Dual polarization LEKID's

  • SOI and 160 um Si wafer fab processes, fabricated at JPL.
  • Mesured responsivity vs pol and measured noise -- looks good.

MC MKID's (along with Kent, Dale, Sherry, Jeff, Rahul)

  • Design has 90% transmission in sims.
  • Using commercially available readout hardware using ROACH1.
  • First measurements coming this summer.


Steve: What do you really need for S4 to buy into this techology? Brad: We need on sky demonstration. Kent: What about lower frequencies? Brad: yes, possible, need low temps, but can achieve this with DR and ADR.

Erik Shirokoff: KIDS

  • Pros/cons to TES vs direct absorption LeKID's vs antenna and horn coupled KID's
  • Working on super spec R = 100 single chip spectrometer using KID's.
  • Noise PSD's look encouraging, but haven't demonstrated low freq 1/f performance yet.

What does on-sky demo mean? Unlikely that we'll get ground based KID CMB experiment in the next 5 years.

John K.: You can swap in a 'guest' array on Keck/BICEP to get a head-to-head on-sky omparison to TES arrays.

Clarence Chang: ANL Fabrication capabilities

  • Working on SPT3G wafers, 271 pixels, 6 datectors per pixel
  • Bringing fab time to < 3 weeks per 10 wafer iteration
  • Will turn attention to S4 fab after 3G is delivered.

Natalie: What kind of yield are you getting? Clarence: 5 wafers survived, but Tc's not foldable, but working TES's per wafer yield is > 90%. Keith: Could you envision 3-10 wafers per day? If not, can S4 production be commercialized? Clarence: Superconducting fab sq ft at ANL is bigger than commercial. Steve: Throughput is not the issue, it's R&D. Only need a couple of hundred foldable wafers in the end.

Toki Suzuki: LBNL/Berkeley detector development

File:2016 03 09 LBNL CMBS4 Hardware.pdf


  • Need to ramp up industrial-sized production for 500,000 detectors.
  • Tried commercial foundry fab for simplified version of CMB detector with optical active antenna coupled TES bloomer, integrated L's and C's.
  • Tested this in test cryostat, and it was optically active.
  • This could be viable path for S4
  • Also working on R&D fab at Berkeley.
  • Automating probes, wire bonding.

Amy: What was cost for comm fab? Toki: $3k for one wafer, but for simplified design.

  • Scalable AR coating using alumina thermal spray. Can use for lenslets and lenses.
  • Demonstrating high yield arrays for S3 experiments (PB/SA)
  • Working on continuously rotating HWP.

Mike: How does off-foresight rays interact with AR coating? Kam: Gaussian beam wavefront is flat at top of lenslets.

Jeff McMahon: Metamaterial AR coatings and HWP's

  • Making metamaterials using Silicon dicing saw.
  • Had to build 3-axis dicing saw in house.
  • Have made a 3-layer coating prototype. Measured reflection < 1% 90 - 270 GHz
  • Have scaled to higher frequencies, 150/220 lenses for AdvACT, defects near zero.
  • Measured reflection ~0.3% over entire bandwidth.
  • Made 5 layer prototype, 75-300 GHz bandwidth. Yet to be measured.
  • Making metamaterial HWP.
  • Need to use high quality Si, and delta ~ 1e-4 at 300 K.
  • Have deployed some HWP's, measured a few K loading.

Jay Austermann: NIST Fab efforts

  • Making detectors, feed horns, readout (SQUIDs for TDM, FDM).
  • Made first MC on-sky array, 90/150 for ACTpol
  • Now making 6-inch wafers for AdvACT, so far looks good.
  • Using unit rhombus design, to avoid extra lithography step for wiring.
  • Very uniform Tc's across 6 inch array (see Dale's paper).
  • AlMn tunable Tc's.
  • Using microwave cross-unders for simplification.
  • Using SiN dielectric for lower loss.
  • Repeatable and well matched spectra
  • Metalized stacked Si feed horns have systematics advantages (low sidelobes, no AR coating needed).
  • Making KID's for BLAST, and CMB. Measuing photon noise limited PSD's for BLAST detectors. Bands look good.
  • Scaling to low frequencies.
  • Also going ton antenna coupled arrays.
  • Readout: microwave mux, ROACH based for MKID's

Nils: Is there a mechanism for NIST to be funded for S4 detector fab? Natalie: Yes, there is a mechanism. Adrian: Talked about this with Kent -- if there is a will there is a way, but depends on scale of effort.

Adrian Lee et al: Strawperson instrument configuration for mapping speed/forcast simulations:

Apertures to consider: Small (0.5 m), Medium (4-6 m), Large (10 m class). By aperture, mean illuminated aperture.

Configuration SM:

Large sky area survey: Four 6-m, 400,000 detectors Inflation survey: 2 x 6m with 200,000 detectors, 20 x 0.5 m with 200,000 detectors.

Configuration SML:

2 x 10 m 2 x 5 m ...

Jeff: Let's not assume a specific configuration, let's give basic parameter ranges to simulators, and let them dial the knobs. Adrian: But, we can go in both directions -- pose a specific config and iterate with simulators. Brad: Agree with Jeff. Adrian: Wu et al paper is a good framework. Mike: Take Wu et al paper framework and vary 1/f knee parameter

Jeff/Brad/Nils: Keep parameters basic: Aperture size(s), number of detectors at each frequency per aperture type, frequency bands.

Clem: But this will push towards no small apertures -- need additional parameter to reflect difficulty for large telescopes measuring low ell: 1/f, scan speed, systematics.

John K.: Can separate by survey goals: two surveys, small sky inflationary B-mode survey, large sky lensing B-modes survey. Optimize each separately.

Adrian: Can tune NET vs frequency spec Akito and Jeff: This is too specific for now.

Clem: Need to set low l limit for large apertures to give to forecasters.

Jeff: Should we agree that low-ell min of 100 for large aperture telescopes?

Clem: Low l min is based on systematics, not 1/f an scanning speed.

Clem: low l limit should be 150 or 200, not 100 for large apertures.

To be continued ...