Tuesday, August 18, 2015

Progress in solid-state memories

Last week's Storage Valley Supper Club provided an update on developments in solid state memories.

First, the incumbent technology, planar flash, has reached the end of its development path at the 15nm generation. Planar flash will continue to be the majority of flash bits shipped through 2018, but the current generation is the last.

Second, all the major flash manufacturers are now shipping 3D flash, the replacement for planar. Stacking the cells vertically provides much greater density; the cost is a much more complex manufacturing process and, at least until the process is refined, much lower yields. This has led to much skepticism about the economics of 3D flash, but it turns out that the picture isn't as bad as it appeared. The reason is, in a sense, depressing.

It always important to remember that, at bottom, digital storage media are analog. Because 3D flash is much denser, there are a lot more cells. Because of the complexity of the manufacturing process, the quality of each cell is much worse. But because there are many more cells, the impact of the worse quality is reduced. More flash controller intelligence adapting to the poor quality or even non-functionality of the individual cells, and more of the cells used for error correction, mean that 3D flash can survive lower yields of fully functional cells.

The advent of 3D means that flash prices, which had stabilized, will resume their gradual decrease. But anyone hoping that 3D will cause a massive drop will be disappointed.

Third, the post-flash solid state technologies such as Phase Change Memory (PCM) are increasingly real but, as expected, they are aiming at the expensive, high-performance end of the market. HGST has demonstrated a:
PCM SSD with less than two microseconds round-trip access latency for 512B reads, and throughput exceeding 3.5 GB/s for 2KB block sizes.
which, despite the near-DRAM performance, draws very little power.

But the big announcement was Intel/Micron's 3D XPoint. They are very cagey about the details, but it is a resistive memory technology that is 1000 times faster than NAND, 1000 times the endurance, and 100 times denser. They see the technology initially being deployed, as shown in the graph, as an ultra-fast but non-volatile layer between DRAM and flash, but it clearly has greater potential once it gets down the price curve.

Thursday, August 13, 2015

Authors breeding like rabbits

The Wall Street Journal points to another problem with the current system of academic publishing with an article entitled How Many Scientists Does It Take to Write a Paper? Apparently, Thousands:
In less than a decade, Dr. Aad, who lives in Marseilles, France, has appeared as the lead author on 458 scientific papers. Nobody knows just how many scientists it may take to screw in a light bulb, but it took 5,154 researchers to write one physics paper earlier this year—likely a record—and Dr. Aad led the list.

His scientific renown is a tribute to alphabetical order.
The article includes this amazing graph from Thompson-Reusters, showing the spectacular rise in papers with enough authors that their names had to reflect alphabetical order rather than their contribution to the research. And the problem is spreading:
“The challenges are quite substantial,” said Marica McNutt, editor in chief of the journal Science. “The average number of authors even on a typical paper has doubled.”
Of course, it is true that in some fields doing any significant research requires a large team, and that some means of assigning credit to team members is necessary. But doing so by adding their names to an alphabetized list of authors on the paper describing the results has become an ineffective way of doing the job. If each author gets 1/5154 of the credit for a good paper it is hardly worth having compared to the whole credit for a single-author bad paper. If each of the 5154 authors gets full credit, the paper generates 5145 times as much credit as it is due.  And if the list is alphabetized but is treated as reflecting contribution, Dr. Aad is a big winner.

How long before the first paper is published with more authors than words?

Tuesday, August 11, 2015

Patents considered harmful

Although at last count I'm a named inventor on at least a couple of dozen US patents, I've long believed that the operation of the patent system, like the copyright system, is profoundly counter-productive. Since "reform" of these systems is inevitably hijacked by intellectual property interests, I believe that at least the patent system, if not both, should be completely abolished. The idea that an infinite supply of low-cost, government enforced monopolies is in the public interest is absurd on its face. Below the fold, some support for my position.