Semiconductor technology

I have four stocks in this category; three I’m pretty happy with, one is a “meh”.

Intel is my current favorite, ticker symbol INTC. One of the pieces of advice I had gotten was “find the clear leader in an industry, buy that, and hold on”. Intel seems to me, to clearly be the leader in semiconductor fabrication technology. I bought it at $24 per share, and it is currently at $47.

I do like that Intel pays a dividend.

Another thing that I liked about Intel is that they had a partnership with Micron, on a type of memory they named “Optane”. I know that everything in computer technology is about the pipeline of storage into the registers of the CPU. If we could make the CPU have enough storage, and, we wouldn’t need external storage, and everything would be going at the full speed of the CPU.

But that isn’t physically possible, if only because once in a while, the power goes out. CPUs use Dynamic RAM (it is the registers the CPU manipulates, and on-board memory called L1 cache). Dynamic RAM is dynamically refreshed with electrical power. When the power drains out, so does the data. Some sort of storage is needed, that doesn’t lose it’s data when the power is off. Since the 1950’s, the “storage” has been external to the CPU, and is orders of magnitude slower than the CPU itself.

I think the Optane idea could (potentially) flip computing on it’s head: the memory becomes so fast, that the pipeline of storage into the registers (and back), can be made direct. Or put another way, the CPU could run at the speed of memory – which is the storage. What if the external storage was the same speed (not orders of magnitude slower) as the CPU? What if the RAM was the disk? What if every register retrieve and store were permanent?*

Now really, even Optane memory does not run at the 2.x or 3.5 GHz of a CPU. Most Dynamic RAM access is in the 1.2 GHz range. So most modern computers spend a lot of hardware design on fetching data from the comparatively slow RAM, keeping as much of it on the CPU chip as possible, and then dealing with cache misses, and branch prediction misses, and all sorts of work to keep things in sync when the whole scheme isn’t perfect.

But what if 1.2 GHz was fast enough? Could it be fast enough, if there was no difference between RAM and storage? If the RAM addresses were the storage addresses?

Optane memory is essentially the next wave of solid state disk; and has capacities of same. How does the game change, when your 2 Terabyte Optane storage means that really, you have 2 TB of RAM? In six years, it will be 16 TB of RAM; eight years = 32 TB, ten years = 64 (if not 128) TB of RAM.

I expect that ten years from now, the Optane memory will have CPU electronics on the Optane chips, and the computing will be done on the memory chips. It’s a lot of work to ship bits off chip to a CPU, have the CPU alter them, and then ship those bits back, across the backplane, to end up back on the storage chips. It’s time consuming, too.

This is what I mean by turning computing on it’s head.

Anyway, it’s probably obvious that I’m a fan, so I like both Intel and Micron Technology – ticker symbol MU. I bought MU at about the same price as it is today. However, six months ago, it was double what it is today. I should have sold 1/2 my position then, and made a note somewhere that what was left is now free money.

*Not “forever” permanent, but from the point of view that “if the power goes out, we don’t care, because the data has already been written to storage”.