2
\$\begingroup\$

Let's say I had a very large static dataset that I'd like to distribute via solid state memory to a lot of people. Let's say it's on the order of terabytes. What would be the densest type of storage I could use?

I read somewhere that in Intel's 14nm process they can fit ~37 million transistors on a square millimeter which would mean that about 400 MB could be stored on a square centimeter. This assumes one transistor per bit, which I assume is as good as you can get for ROM. (is this a good assumption?)

But then I notice that you can buy a 256 GB microSD card which is smaller than my hypothetical ROM. How the hell is the flash so dense and would it be possible to make a ROM that dense (or denser)?

\$\endgroup\$
8
  • 1
    \$\begingroup\$ How do you translate transistors into MB? \$\endgroup\$
    – Eugene Sh.
    Commented May 16, 2017 at 16:46
  • \$\begingroup\$ Silicon pales in comparison to the 215 x10^15 bits per gram that DNA offers, but silicon is currently faster. \$\endgroup\$
    – glen_geek
    Commented May 16, 2017 at 16:57
  • \$\begingroup\$ I know it's 6 transistors per bit of SRAM and 2 transistors per bit of SDRAM. Don't know the figures for ROM and Flash...1 transistor sounds sensible: just the sense transistor. \$\endgroup\$
    – TonyM
    Commented May 16, 2017 at 16:59
  • \$\begingroup\$ @glen_geek, can people now manufacture DNA memory in volume then? \$\endgroup\$
    – TonyM
    Commented May 16, 2017 at 17:00
  • \$\begingroup\$ @TonyM men do fairly often, but mostly it just gets wasted \$\endgroup\$
    – BeB00
    Commented May 16, 2017 at 17:05

1 Answer 1

2
\$\begingroup\$

There are a variety of ways of encoding more than one bit per transistor by having each transistor control an "analog" signal. That could be accomplished in a ROM, albeit at considerable cost, if one used a process that could lay down 15 slightly-different kinds of transistors that would pass different amounts of current. To read a row of bits, apply a current source to each column, turn on the row, and then measure the resulting voltage on every column.

Flash can actually operate on a somewhat similar principle, but with a twist: rather than having to manufacture transistors differently, the flash effectively applies weird voltage conditions to them in such a way as to implant or remove electrons to/from areas that are generally non-conductive. A major advantage of that approach versus one based on manufacturing different kinds of transistors is that it's possible to hit a transistor with a little bit of charge, measure whether it has received enough yet, hit it with a little more, etc. until each area has received the perfect amount of charge. Doing that is much easier and cheaper than having to tweak manufacturing processes to produce 15 different kinds of transistors directly.

It should be noted that while efficiently manufacturing ROM using multi-level techniques is harder than using such techniques in flash, ROM does have an advantage in some applications: ionizing radiation that hits a ROM while it is being accessed might cause a bit to be misread, but it will not alter the physical arrangement of transistors on the chip. By contrast, if information is stored by the distribution of charges on a chip, ionizing radiation may permanently change the content of such stored information. Consequently, manufactured ROM may be advantageous in contexts that need operate in high-radiation environments.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.