Name:
Email:
Do we have smearing in rolling mode and can we combine rolling mode with vertical hardware binning?
(a) Distortion should be minimal in rolling shutter mode for most applications. It is common practise to take images at a rate that is not going to result in major movement of object within a frame. Typically expect objects to move sub-pixel during the time for rolling shutter readout front to move from top to bottom of object. (b) There is no full vertical binning allowed, since binnign is not on chip. Thus the noise from all pixels woudl combine in quadrature.
Are there any tradeoffs when using the Spurious Noise Filter?
Not really, it is discriminate enough to only work on spurious background events and leave signal alone. It is done in FPGA so shouldn’t affect speed either. We kept it optional though as know it will make some uncomfortable to filter high noise pixels, even though it will only filter them if no signal falling in them.
At the peak of its QE, is sCMOS camera capable of doing single photon imaging?
No, still need an EMCCD for single photon counting. A single photoelectron would be lost within 1e rms read noise.
Can I reduce FOV in experiments? Will that be useful to increase frame rate?
Yes, simple to select between multiple ROI choices and it will increase frame rate proportional to decrease in vertical dimension.
Can the sCMOS camera be binned before readout?
sCMOS is binned after read, so 2x2 binning will increase noise by x2 (note, not by x4)
Can the sCMOS replace ICCD (iStar) in LIBS experiments? Can you attach an intensifier in the sCMOS? Can you provide minimum delay and integration times of the camera?
To an extent it will be used for LIBS. The transfer time into the readout node will be ~ 1us which should work OK for a lot of LIBS.
Could you please again explain why 2 readouts are required for global shutter?
Basically in rolling shutter charge transfer happens on a per row basis whilst in global shutter charge transfer happens for the whole sensor or globally. Typically to readout a pixel the following has to happen:
1. Reset the readout node2. Measure the node level (reference level)3. Transfer charge from photodiode to node4. Measure the node level (signal level)5. Subtract 2 from 4 to get real signal.
This process is commonly referred to as CDS (correlated double sampling) and is done in the analogue domain before digitisation. If the time between 2 and 4 becomes too large then drifts in the node level produce significant errors in the real signal and hence increase the ‘noise’ of the measurement.
For rolling shutter charge transfer happens on a per row basis, therefore for each row the node is reset, then read (reference), then charge transferred, then the read again (signal), thus the reference and signal measurements are close together (< 10 usecs @ 280MHz) and so there is little time for voltage drift in the node and an increase in ‘noise’. Then the next row is done, etc until the whole sensor is readout. The disadvantage of this is that the start and end exposure time moves by the row readout time for each subsequent row. So whilst each row of pixels is exposed for exactly the same length¬¬ of time they do not all start and end at exactly the same time.
In global shutter, because we want the start and end of the exposure to occur at exactly the same time for every pixel (not just for pixels in the same row) point 3 above has to happen for all the pixels at the same time. However because of this the reset and reference read would have to have occurred before this global transfer for every row, and then we would have to do the signal read. For the fastest readout rate (280MHz) this would be mean that there would be a minimum of 10 usecs (time to measure a single row) x the number of rows in half the sensor between the reference read and the signal read which would allow for drift in the node level and thus significantly increased noise. In order to avoid this a reference frame is actually digitised and then the signal is digitised and the two subtracted to get the ‘real signal’.
There is a little bit more involved than this but essentially this explains the reasoning behind why there are two images for global and only one for rolling.
Data readout - if 100 fps is possible at 5.5 MP, is a higher frame rate possible at 1MP ROI?
Absolutely yes. Please see Neo spec sheet were have supplied a max frame rate table vs ROI
Is it water cooled? So, no fan, no vibrations? Is that standard or specific to slide #31?
The Neo can be fan cooled to -30C and water cooled to -40C. If water cooling, can turn off fan to reduce vibrations.
Do I understand correctly that for low light intensity EMCCD is still a preferred option?
Yes, for extremely low light its hard to beat a back-illuminated 'zero noise' EMCCD. I would estimate the majority of single molecule detection will still be better off with EMCCD and a lot of cell microscopy will find more benefit from Neo. I'm generalising of course.
Do readout times for ROIs scale as the number of pixels or are rows and columns asymmetric?
Frame rates with ROIs scale with saving in the vertical direction. Each row must still be read out. If you can use elongated 'letterbox' ROIs, you may find some advantage in doing so. Not however that if spooling to hard drive there will be some future benefit to selectign ROIs that sacrifice in the horizontal also, in terms of reducing file size during transfer, thus allowing faster spool frame rates.
Do you see this camera having applications in spectroscopy or is it primarily for imaging?
For echelle spectroscopy possibly, but beyond that vertical binning is generally not recommended as it will incorporate noise from each pixel binned.
Slide 25 - does a single image include a mix of pixels with different Hi or Lo gain regimens?
Yes, that’s what happens. A choice is made based on the signal intensity per pixel as to which amplifier signal to use.
Does it have a fire output?
Yes.
Does the sCMOS camera have a double exposure option (PIV mode), which means it can capture 2 succesives images with micro second or less dt between the 2 images?
It has been tested this way using global shutter mode. Time between images right now is ~ 2us.
Fixed pattern correction - it seems like you only need to set a correction for every column - or are you correcting indeed for all pixels separately (5.5M values per table)
We have correction for every pixel. It is true that column offset variance is th egreatest contribution but there is residual noise left at the pixel level. Better to do it thoroughly.
For real-time analysis, I do not want to buffer the frames but read them out continuously. Can this be done, at full speed, full frame and 16-bit dynamic range?
16 bit file size for the 5.5 MP image during transfer is 11.2MB, so 100fps is 1.12 GB/sec. Note that even dual camera link (when we activate it in future) will not have enough bandwidth to continuously spool 16-bit at this rate, would need to drop to 11-bit to achieve, or alternatively slow down a little. However, the 4GB on head memory can burst 16-bit data at max speed now.While all data passes through the 4GB buffer, it doesn't impact the ability to spool and do real time analysis.
Given the performance of sCMOS, what applications are "conventional" iXon CCD cameras best suited for?
The extremely low light applications will still benefit from iXon 'zero noise' raw sensitivity, e.g. single molecule detection and single photon counting.
Can you estimate the additional effect of high energetic cosmic ray events to the noise-floor? Is this effect lower than in deep cooled CCD-Systems? Also, you compare the noise of a 60% QE sCMOS with a 90%+ QE EMCCD. Do you have compared with a 60% EMCCD and what is the difference?
You may experience the occasional cosmic ray under long exposures, just as for CCDs. I havent yet proven, but reckon the spurious noise filter would be capable of flattenign these events if they occur. Yes, I've compared to the 'virtual phase' EMCCD that has ~ 60% QE and 8 um pixels. In this case I assumed that there is no binning of sCMOS or interline as they all have relatively small pixels. There is a cross-over point at ~ 12 electrons per pixel, below which EMCCD would give better signal to noise. Compare this to cross-over point of ~ 22 photons/pixel in case of back-illuminated EMCCD.
Why is a pixel size of 6.5um is better than 8um?
6.5 um has proven a good option for Sony interline over the years, thinkign primarily of microscopy. Is good size to oversample the diffraction limit of a x60 objective and give sharper resolution of fine intracelular structure.
How can we get information on the number of photons per pixel with this camera?
The performance sheet provides information on number of electrons per count for chosen gain sensitivity selection. You will also know the baseline bias offset. First subract the baseline, then multiply by the gain sensitivity in electrons per count. That will bring you to number of electrons in each pixel. To estimate incident photons per pixel, would then divide by QE at your wavelength, as a fraction, e.g. if QE was 40%, divide by 0.4.
How do you implement the dynamic gain output for the pixels into your image date for being properly able to compare the intensities (i.e. real photon numbers)? Is that done on the fly as well?
The signal from the low gain pixels are accuractely multiplied by the factor sensitivity difference between low and high gain, as measured for each camera. That gives linear continuity in counts between pixels sampled by either amplifier. This is done on the fly in FPGA.
How does Neo sCMOS deal with very bright sources? What happens to the extra electrons?
The sensor has x10,000 antiblooming to prevent over saturation of pixels.
How would sCMOS compare in a spinning disc microscope such as Olympus DSU, not using laser illumination, when compared with EMCCD?
Probably very favourably as I think signal would typically be in a regime where SNR ratio of each detector would be comparable. In plot from slide 11 of webinar, this refers to > 22 photons incident per 13 um pixel area (2x2 binned in case of sCMOS).
If a pixel becomes saturated is there any electronic control of possible local blooming effects?
The pixels have anti-blooming drains. Should allow for up to x10,000 illumination of the well depth.
If I understood well, the pixels can have different gains. If I am interested in the counts of photons themselves, not in the "image", as in astronomy when we are interested in sum counts in a given area of the sensor, I suppose it can have some caveats for astronomy work, for instance. Am I right?
Is it the dual amplifier architecture you refer to? The signal from the low gain pixels are multiplied by the factor sensitivity difference between low and high gain. That gives linear continuity in counts between pixels sampled by either amplifier. This is done on the fly in FPGA. The camera is very much designed to be quantitatively reliable.
In order to achieve maximum frame rate are there unique requirements to the dimensions of the ROI? Specifically does the ROI have to be small in a single dimension (example 1800 X 8 pixels)?
Yes, you gain much more in speed by reducing the ROI in the vertical direction, since the entire 2560 pixel width of each row must always be read out.
In SNR comparison plot, was there any normalization for differences in QE? If noise increases by 1.4 in EMCCD due to amplification gain, I would still expect EMCCD to perform better at higher light levels [0.9 > 0.6 x 1.4 = 0.84].
Yes, QE differences were factored into these plots. From a signal to noise perspective, multiplicaiton noise has the effect of halving the QE. Therefore the 90% of back-illumination actually effectively becomes 45%. So its more like comparing 0.6 sCMOS vs 0.45 EMCCD.
Is the dual-amplifier mode optional, for precise whole-image quantatitive analysis?
Yes is optional, selectabel within the software. You can still opt to readout single amplifiers in 11-bit, for reduced file size and faster frame spooling rates.
Is the offset compensation automatic and standard or one has to set it via software?
It's automatic, don’t need to set anything the camera detects your configuration and applies appropriate compensation data.
Is there a low limit to the Neo sCMOS fps?
Can set up mutliple hour exposures if desire. But need to be careful of dark noise build up. Would actually recommend interline CCD (Clara with lwo noise 1MHz mode) to give better SNR under very long exposures, greater than 1 minute.
Is there any commercial CMOS camera that can support frame rates bigger than 100fps or Neo sCMOS is the fastest CMOS in the market now?
There are specific high speed CMOS cameras, but generally very high read noise and used for apps where illumination is no problem, e.g. slow motion photography, crash test etc.
Is there any condensation on the window if the sensor is cooled at -40 degrees?
No, there will be no moisture on the window, the Neo uses a vacuum sensor enclosure so there is no convection current to cool window.
Seems like this would also help for fast Calcium imaging?
It should be a great camera for fast calcium imaging, but like any detector, just need to be careful you have sufficient signal to provide enough photons per pixel during each short exposure of the series, otherwise signal to shot noise ratio would suffer.
So the buffer is good for 4 seconds?
Should be closer to 5 seconds in reality if you have a relatively fast hard drive - as soon as data enters buffer it is spooled to hard drive, therefore should effectively achieve greater than 4GB data flow capacity before it becomes filled.
What's the disadvantage of the rolling shutter?
In reality very little disadvantage. In some applications where larger objects are moving fast across the field of view (and in those circumstances expect some smearing anyhow, irrespective of readout type, due to movement of object during a single exposure) then might expect some level of spatial distortion due to time taken for 'readout front' to transverse from bottom to top of object while object is moving. However, for most applications, expect the object to move sub-pixel distances during this time to transverse, resulting in negligible spatial distortion.
When will you offer 8, 16 or even 32 GByte on-head memory as an option?
We won't go this route. We decided 4GB was enough to give us a some burst capability and, just as importantly, to provide sufficient memory buffer for advanced FPGA processing while also protecting the acquisition series from Windows interrupts. It costs a lot more for camera memory than for PC memory, so instead when we activate the 2nd cameralink port we will recommend PCs with RAM for extended bursts at full speed.
Why is aquisition done in 11 / 16 bit? why not 8 / 16?
The ADCs at the end of each column are 11-bit. We didn’t want to use 8-bit as some users will be happy to only use one amplifier/ADC rather than the dual amplifier mechansim for 16-bit. In this single amplifier mode, better to have 11-bits to play with though rather than 8-bits.
Will the data flow monitor calibrate a RAID array?
Yes, the data flow monitor will calibrate a RAID array.
Will the noise performance of 1.4 electrons remain same whether operated in global shutter mode or rolling shutter mode?
The noise in global shutter will increase by factor of root 2 (1.41). This is due to the additional reference read that is required for global shutter readout. Note two reads does not double noise, noise increases by 'root of sum or squares'.
How fast is the shutter?
Takes 2us to transfer signal into readout node.