Hacker News Reader

Comment by lifeformed | Hacker News Reader

> ultra-fast

Does it say how fast the read/write speed is?

They are getting 12 kbit/s right now, and claim they could get Mbit/s to Gbit/s using very fast light modulators (which themselves are research-lab projects, to my knowledge).[1]

The phrase "ultra-fast" laser is a term of art, which really means "laser with ultra-short pulses". They are using a laser with pretty short, 280 fs, pulses, but only a 200 kHz repetition rate.[2] It evidently takes several pulses to write one bit. Due to the physics of femtosecond lasers, you can't easily increase the rep rate without compromising the pulse length and intensity, which both need to be very good in this application.

The readout "should" be faster than "conventional" methods, but no details are given.[1] I don't see evidence that they have shown even moderate speed readout. It looks like they took micrographs and then inspected the images to determine the values of the bits.

[1] Gizmag article referenced in another comment, http://www.gizmag.com/superman-memory-crystal/28231/?utm_sou...

[2] Authors' original paper, http://www.orc.soton.ac.uk/fileadmin/downloads/5D_Data_Stora...

A "balanced" computer system used to follow the "one second" heuristic. (Is this named after somebody?) That is you should be able to offload temporary memory (core) to permanent storage in one second. And it should be able to execute a computation on every element of memory in one second too. I know this has held up through the gigaword generation. Leading edge computers, like a supercomputer, may be unbalanced where one of these factors lags.

Seymour Cray followed this balance of processor-memory-bandwidth to memory-bandwidth to storage in his designs.

Let me correct myself. I came across Jim Grey's 1999 paper "Rules of Thumb in Data Engineering" where he states that this heuristic is from Gene Amdahl in about 1965: a balanced system has one bit of I/O per second for each instruction per second and one byte of memory for each instruction per second. So, 8 MIPS per MBpsIO, and one MB/MIPS.

I believe the article mentioned femtosecond laser writing. I don't know how much data gets written in each pulse though.

"The experiments were performed with a femtosecond laser system Pharos (Light Conversion Ltd.) operating at 1030 nm and delivering 8 μJ pulses of 280 fs at 200 kHz repetition rate."

My €0.02 bet is that that means 200 kilobit per second. In layman's terms: about twice as fast as we could send data to Voyager 1 when it was near Jupiter aka 'abysmal'.

But use a hundred lasers in parallel and you're up to 20Mb/s!

"delivering 8 μJ pulses of 280 fs at 200 kHz repetition rate."
8.0E-6 J * 2.0E5/s = 1.6 J/s = 1.6W
So, this laser outputs 1.6W of power (in very short pulses, but that is irrelevant here). Let's assume a LED powered laser. http://en.wikipedia.org/wiki/Energy_conversion_efficiency gives, optimistically, 35% efficiency for a LED. That would mean a single-laser version would use 5W, a 100-laser one 500W of input power.
=> Some improvements are needed before this replaces hard disks at home. We'll see whether there is room for that.

Reply on news.ycombinator.com