Open in TheBrain

WU, Ming, MEMS Photonics, CAL, Jan 11, 2016

<p class="MsoNormal" >WU, Ming,&nbsp; MEMS Photonics,&nbsp; CAL, Jan 11, 2016</p>

<p>

he is a prof at CAL, Berkeley.</p>

<p>

This is at 4:15PM&nbsp; on Monday in Spilker 232&nbsp; (Applied Physics) at Stfd.</p>

<p>

He jokes... everytime he is here&nbsp; CAL wins the Big Game&nbsp; (vs Stfd)</p>

<p>

he begins...</p>

<p>

consider web traffic to Facebook alone....

1.45 B&nbsp; monthly users&nbsp;&nbsp; growth is 13% per year;

most users are&nbsp; ages 18 to 25...

1.25&nbsp; mobile users</p>

<p>

he shows mega data centers</p>

<p>

<b>every 60 seconds&nbsp; 204&nbsp; M emails</b>

<b>5M&nbsp; google searches.</b></p>

<p>

<b>it may&nbsp; take 500 machines to process a single inquiry&nbsp; (in Google Cloud)</b></p>

<p>

what is a data center...?</p>

<p>

shows a slide by&nbsp; Laurent Schares&nbsp; at IBM...</p>

<p>

<b>largest may have&nbsp; 1 million sq feet and 100,000 servers....</b></p>

<p>

servers connect to top of rack (TOR) switches...electrical...</p>

<p>

he compares&nbsp; EPS = electrical packet switch&nbsp; vs&nbsp; OCS&nbsp; = optical circuit switch...</p>

<p>

at 10G and 40Gbit speeds&nbsp; it is electrical&nbsp;&nbsp; but gets challenging when faster...</p>

<p>

TOR = top of rack&nbsp; switch... to the host servers...</p>

<p>

want to have a CORE that&nbsp; connects to the various TORs...</p>

<p>

HP Lab&nbsp; OmniSwitch...&nbsp; change tree structure on the fly...</p>

<p>

available now&nbsp; is&nbsp; 3D&nbsp; MEMS Switch&nbsp; (his focus today)...</p>

<p>

but now it&nbsp; is a free space switch...&nbsp; CALIENT&nbsp; is one company&nbsp; eg...</p>

<p>

320 by 320&nbsp; connections... only 1.5 dB loss...per connex.</p>

<p>

but cost is abt&nbsp; $100 per port...</p>

<p>

so&nbsp; 320 by 320&nbsp; =&nbsp;&nbsp; 30 to 50 thousand dollars for a switch...</p>

<p>

<b>but we need at 10 dollar per port switch for data.</b></p>

<p>

and need thousands of ports and faster switch time

currently s at&nbsp; 10 to 20 ms&nbsp; but really want&nbsp;&nbsp; 10 to 100 microsec&nbsp; switch.

<b>but really want&nbsp;&nbsp; < one microsec.</b></p>

<p>

and low insertion loss&nbsp;&nbsp; ideally < 2 dB.</p>

<p>

so&nbsp; how abt&nbsp; <b>silicon&nbsp; photonics...</b></p>

<p>

current photonic switch is free space based...</p>

<p>

but also&nbsp; NTT has made&nbsp; 32 by 32&nbsp; switch&nbsp; on planar lightwave circuit

&nbsp; but its huge&nbsp; abt&nbsp; 4 inch chip...</p>

<p>

silicon photonics... can do billions of&nbsp; transistors....

commercial products available&nbsp; BUT&nbsp; <b>high optical loss and only small port counts...</b></p>

<p>

can get&nbsp; good switching in abt 1.3 microns&nbsp; to get crossover into&nbsp; two waveguides...</p>

<p>

confinement of light to silicon waveguide.</p>

<p>

shows&nbsp; PILOSS switch...&nbsp;&nbsp; and each node has&nbsp; abt&nbsp; 16 dB loss.</p>

<p>

path-indep insertion loss&nbsp; = PILOSS.</p>

<p>

everybody is at&nbsp; 0.5 dB per port...</p>

<p>

only two ways to do&nbsp; switching...</p>

<p>

change refrac index&nbsp; by injecting holes into Si.

to modulate carrier densty&nbsp;

VS&nbsp; thermo-optic effect&nbsp; (eg&nbsp; by&nbsp; heating it up by hundreds of degrees.</p>

<p>

both are very small&nbsp; effects...</p>

<p>

so v inefficient...</p>

<p>

<b>want to use&nbsp; MEMS instead&nbsp; to get 100% index modulation</b>&nbsp; (instead of 1% above)</p>

<p>

no cross talk...</p>

<p>

<b>with mems&nbsp; ALMOST NO&nbsp; OPTICAL LOSS</b>... ONLY&nbsp; -.01 dB at cross points...</p>

<p>

our # of crossing is only&nbsp; 2N -1&nbsp; .</p>

<p>

not available as a product... <b>our students make these</b></p>

<p>

<b>shows suspended waveguide&nbsp; which is&nbsp; 220 by 400 nm...</b></p>

<p>

<b>uses&nbsp; MEMS cantilever.</b></p>

<p>

but we have a big loss... so, what went wrong??</p>

<p>

better to use&nbsp; wider waveguides...</p>

<p>

with flapping structures...&nbsp; (like&nbsp; an LA freeway on the 405...

we just mechanically&nbsp; lower down the entire&nbsp; freeway overpasses)</p>

<p>

<b>like a flapping birds wings...</b></p>

<p>

<b>using adiabatic couplers...</b></p>

<p>

shows a video animation...&nbsp; as orthogonal flaps come down&nbsp; the light goes thru it at right angle

instead of&nbsp; straight thru.</p>

<p>

uses a mechanical stopper to stop the descent of flaps...</p>

<p>

<b>he shows&nbsp; 64 by 64&nbsp; si photnic switch in 1 cm2.</b></p>

<p>

unit cell size is only 110 microns sq.</p>

<p>

next he shows&nbsp; Gen2 switch...</p>

<p>

gap closing actuatiors...</p>

<p>

takes&nbsp; abt&nbsp; ten volts to switch&nbsp; (if biased in middle of hysteresis loop)</p>

<p>

see SEOK et al&nbsp; in GFP in 2015&nbsp; = group4&nbsp; photonics&nbsp; conference.</p>

<p>

Seok Quack&nbsp; Han, et al.

reliability study...</p>

<p>

it is reliable&nbsp; no degrad at&nbsp; 10B cycles of switching</p>

<p>

andno stiction even afger long contact...</p>

<p>

so how to use this... (what are the apps?)</p>

<p>

multicast switch...&nbsp; to one to many...eg like trad. broadcast&nbsp; (one show eg&nbsp; to N users)</p>

<p>

&nbsp;

&nbsp;</p>

Attachments