DLP650LNIR: High speed DMD for NIR

Yujie Shen

Part Number: DLP650LNIR
Other Parts Discussed in Thread: DLP7000, DLPC410

Hi,

We are thinking of implementing DMD in our laser scanning system. We have a laser close to 1000nm in wavelength so I checked DLP650LNIR first. But we also need the pattern rate to go above 25kHz. So I have two basic questions:

1. The spec indicates that DLP650LNIR could reach a max rate of 12.5kHz, but I'm wondering if we are only updating half of the pixels could we double the pattern rate? If so, could we make the ROI in any shape (as long as it only contains half of the pixels) or it needs to contain an integer number of rows or columns?

2. I believe the DLP7000 chip could reach 30kHz but that's a visible chip. If I use the NIR laser what might the loss be? Is it possible to get a custom coating, or maybe a non-coated version so we could use third-party coating to make it work in the NIR range? If there's a similar application please let me know.

Thanks

over 3 years ago

+1 Fizix over 3 years ago

TI__Guru 54705 points

Hello Yujile,

Yujie Shen said:
1. The spec indicates that DLP650LNIR could reach a max rate of 12.5kHz, but I'm wondering if we are only updating half of the pixels could we double the pattern rate? If so, could we make the ROI in any shape (as long as it only contains half of the pixels) or it needs to contain an integer number of rows or columns?

The speed relationship is not strictly linear. Using half of the pixels will not result in twice the speed. The row load speed does get reduced to 1/2 of the time, but the mirror clocking pulse (MCP) is relatively fixed. If you used phased mode operation (like a scrolling shutter on a cell phone camera). You can collapse some of the MCP time into the loading of the blocks of data. Quad block mode will give you the best advantage.

Please be aware that this DMD only has a 16 bit wide pixel data bus as opposed to the DLP7000 which has a 32 bit wide bus. It will likely not reach 25 kHz, but might be fairly close using 1/2 of the rows in quad block mode. If your implementation requires that the whole pattern be displayed at the same time, then to use phased, you would need to have very short laser pulses that display in the short time after the second quad group is finished settling, while loading the first quad group for the next pattern.

Yujie Shen said:
2. I believe the DLP7000 chip could reach 30kHz but that's a visible chip. If I use the NIR laser what might the loss be? Is it possible to get a custom coating, or maybe a non-coated version so we could use third-party coating to make it work in the NIR range? If there's a similar application please let me know.

The AR visivle coated glass drops to about 80% transmission for one pass, so that window losses alone will account for about 36% loss. This would not include any fill factor and diffraction efficiency losses.

The DLP7000 is not rated for any significant power in the NIR range. Use outside of the Recommended Operating Conditions will void any warrantee on the DMD.

We do not offer any non-coated version of the DMDs.

I hope this helps.

Fizix

0 Yujie Shen over 3 years ago in reply to Fizix

Prodigy 45 points

Hi Fizix

Thank you very much for the info. Just wondering that during operation of DLP650LNIR, exactly how much time would be spent on loading data and how much on the micromirror reset (Sorry there are just too many numbers in the spec sheet, not sure which one to look for)? Also, I guess for our application, the Load4 Mode might actually be what we need (we'll put consecutive 4 or 8 rows in the same state). Wondering if that will give us some significant time advantage?

Another minor question regarding the mechanical spec of the DMD: there's a plus/minus 1 degree variation to the nominal 12 degree deflection angle, I'm wondering if this it means that this angle will vary from pixel to pixel, but for a single pixel the number will be fairly stable over operations? Or does it mean during operation the angle will keep changing?

Thanks

Yujie

0 Fizix over 3 years ago in reply to Yujie Shen

TI__Guru 54705 points

Hello Yujie,

Let me first address the Load4 Mode. This mode loads 4 consecutive rows with identical data. So if individual row data is required, then it will not work. Also it is every 4 rows starting at a multiple of four. If indeed you can use Load4, it will improve your load speed for those rows.

Regarding the loading and Mirror Clocking Pulses (MCP - also called a "Reset"). MCPs are done by block. The data sheet will specifies that the DLP650LNIR DMD has 16 blocks. This means that there are 50 rows per block. Once a block (or group of blocks - i.e. dual, quad, or global) has started an MCP they may not be addressed (written to) for 12 us.

The DLP650LNIR device has row writes that are 40 clocks long (see the DLPC410 data sheet under 8.3.1.7 Supported DMD Bus Sizes on page 28). This means that a row write takes (40 / 400MHz) = 100 ns to write data. Multiply this by the number of rows that you need to write to.

I hope this helps.

Fizix

0 Yujie Shen over 3 years ago in reply to Fizix

Prodigy 45 points

Hi Fizix

Thanks a lot for the lesson. A few more question regarding DLP650LNIR:

1. Yes Load4 should work for us as we do not need individual row data. I'm also wondering in this mode do we need to load the whole chip, or can we choose to load only 8 rows in each block (say row 1-8 in block 1, row 57-64 in block 2, row 105-112 in block 3, and so on)? Will this further enhance the loading speed? Also exactly how much time does the loading cost in this mode (in us)?

2. For resetting, we also only need to address a subset of the block pixels. So if we use quad or global mode for reset and would like to only address 8 rows in each block, will the write time be reduced to 100ns*16*8 = 12.8us? Also I'm a bit puzzled: is the quad block mode faster than the single block mode because the grouped 4 blocks will reset simultaneously?

Thanks

Yujie

0 Fizix over 3 years ago in reply to Yujie Shen

TI__Guru 54705 points

Hello again Yujie,

Typically you do not want to switch back and forth between Load4 and regular mode, but I do believe that you can randomly address the starting row. The row load time is the same since it uses the same data to load all 4 rows, so you are correct that the load time would be only 12.8 us.

Unfortunately if you are only loading 8 rows (or row writes - even if using Load4) per block, then you can load all the data for that pass in the same amount of time that it would take for a reset and settle, so you might as well use a global reset. Therefore you would have 12.8 us load +12 us reset & settle. This would give 24.8 us per cycle. If we round that to 25 us, then the pattern speed would be 40 kHz or slightly faster.

Fizix

0 Yujie Shen over 3 years ago in reply to Fizix

Prodigy 45 points

Hi Fizix

Sorry for the delayed reply, and thanks for confirming---that sounds promising! Just one last question then: say we are using the scheme that only 8 rows per block are updated, and each cycle the DMD would take 12.8us to load, and then 12us to reset and settle, and stay in the state for 15.2us (so total 40us ~ 25kHz). Am I understanding it right that for the real physical movement of the micromirrors will only happen in the 12us reset and settle time, and remain unchanged for the rest 28us in the cycle? Or shall I expect mirror movement in the loading time as well?

Thanks

Yujie

0 Fizix over 3 years ago in reply to Yujie Shen

TI__Guru 54705 points

Hello again Yujie,

You are pretty much correct. However there is some minor movement of pixels that have changed from 1 --> 0. If you are using a very small NA (large f # system) then you may actually see a fluctuation in the output intensity. This is very much dependent on the system optics. A carefully designed optical system would likely be able to mitigate this effect.

Also using coherent monochromatic light will also influence the optical system design.

Fizix

DLP®︎ products

DLP products forum

DLP650LNIR: High speed DMD for NIR