[jyl]

Flash memory for consumer devices (like USB sticks) is moving to TLC (three bits/cell) and smaller geometries, in order to drive cost down. This makes the flash less reliable, fewer write cycles, and requires more sophisticated wear leveling and error correction in the memory controller chip. PC SSDs will go this way too, and eventually server and storage SSDs. I have more faith that a major SSD maker has done the work, than that some no-name Chinese USB stick maker has.

There is an interesting conundrum coming up. SSD technology is on a more steeply declining cost/GB curve than is hard drive (HDD) technology. If this continues ("if") then at some point SSD will be cost competitive with HDD, if you also consider the advantages of SSD (speed, latency, reliability, power, heat, size). High end enterprise SSD already are close enough in cost/GB to high end enterprise HDD to make SSD the better choice for some applications.

Let's suppose that sort practical parity is reached for mass market applications, like PCs. The market should "tip" to SSD, the share of SSD should grow hugely and rapidly at the expense of HDD share. The amount of SSD sold should skyrocket. To replace a modest percentage of HDD share, SSD volume will have to go up 20X or more.

But . . . making NAND is incredibly capital intensive. It is made in an advanced semiconductor fab, each fab costs $5BN or so. Yet the business of making NAND is not very profitable. It is basically like the DRAM business. Memory is a commodity, the margins are low, the business has a very bad return on capital. The leading memory makers are profitable maybe 1/2 the time. The lagging ones are an inch from bankruptcy.

As a result, the industry is not eager to build dozens of $5BN fabs. Currently, the industry has cut investment way back. The leader, Samsung, is even converting memory lines to make other chips.

Something has to change. Either the cost/byte curves changes or the margin structure of the industry changes.

There are interesting implications.