I've been watching storage for some 40 years. And each decade has seen accelerating change, and none more so than the the past 10 years. That trend will continue. Here's the biggest trends I expect to affect all who use storage, whether they know it or not.
Despite Intel's muffed Optane launch, a hurried affair that led to a dramatic reduction in the original specs they quoted, non-volatile RAM has had - and will continue to have - a dramatic effect on system designs. Whle Optane is the best known NVRAM, other NVRAM devices have been shipping for years, embedded in other devices, often as buffers in controllers.
These devices include:
- Magnetoresistive RAM (MRAM) from Everspin, which has been shipping in volume for much of the last decade. They recently announced an OEM customer design qualification for their newest 1 Gb MRAM part.
- Ferroelectric RAM (FRAM), from Cypress. While the density isn't great - which is why it isn't replacing DRAM - the parts are tough, fast, and use very little power. They'll be common in IoT sensors and in automotive applications.
- Carbon NanoTube (CNT) RAM. CNT NVRAM, by Nantero, should be available from a major foundry in 2020. Almost as fast as DRAM, lower cost and power, and requiring no new fab capacity, CNT is the most promising NVRAM for mass adoption in the 20s.
- Even if CNT RAM never comes to market, Intel's Optane is also promising to shake up the DRAM/Flash duopoly for memory and storage. But it seems Intel is dealing with bigger problems than Optane, so it isn't getting the attention it should. Or perhaps it isn't living up to its earlier promise. We'll see.
Data centric computing
As we collect more and more data from IoT, edge devices, and higher resolution sensors of all types, the amount of data we want to access will overwhelm network bandwidth. The only architectural solution: place processors - CPUs, GPUs, neural processors, and more - as peripherals to massive data stores.
Cheap NVRAM would help data centric computing take off, as it removes the need for backup, and would be ideal for streaming data analysis. Intel's Rack Scale Design architecture is one instantiation of data centric computing . Early versions of other types are probably already running, in secret, inside the big cloud vendors.
Computing has always been about the data, but the rapid progress of CPUs, and the relatively small - compared to what's coming - amounts of data we've stored, has kept our attention on the how of computing, rather than the why. Data is the why of computing, and it will come to the forefront in the next decade.
Data mining regulation
Social networks now connect nearly half of the world's population, a percentage that will rise to three quarters or more in 20s. We've started to see the abuses that these unregulated data collectors can support. Criminals, scammers, dictators, and good, old-fashioned, unintended consequences, will lead to many more.
The hands-off policies of the last 30 years have worked. The internet is vibrant, but now it needs rules, and enforcement, to keep it from becoming even more of a toxic waste dump and playground for bad actors.
Self-regulation is not viable, not when there is so much money to be made, or influence to be had. It is time for the people to step in - through their governments - and force surveillance capitalists to meet minimum standards for the ads they accept and the algorithms they use.
The Storage Bits take
Those aren't the only advances we'll see by the end of the next decade. Expect 80TB hard drives, terabit interconnects, and very fast specialized coprocessors, such as multi-tera-op neural processors.
For civilians perhaps the biggest effect of massive data stores will be the social effects of instant recall of everyone's youthful follies. In my generation high school yearbook photos were probably the most embarrassing relic of one's youth. Now, social media preserves video evidence of your half-formed brain.
Perhaps that will lead this generation to be more tolerant and less judgmental. If only.