say farewell to reassuringly boring industrial SSDs

this is not your Grandfather's industrial SSD market

• the ability to support ultra low power (no battery) rugged computing nodes.
  
  Let me begin by asserting that FPGA-style rich functionality IP libraries for customer printable graphene nvms and SSDs are still many years in the future...
  
  You don't have to take my word for it.
  
  You can judge progress for yourself by seeing where that technology is now (2017) - Multilevel Ultrafast Flexible Nanoscale Nonvolatile Hybrid Graphene Oxide–Titanium Oxide Memories compared to the state of the art 7 years ago in this (2010) paper - Graphene Oxide Thin Films for Flexible Nonvolatile Memory Applications (pdf)...
  
  The pace of that productization clock looks similar to the timeline 14 years of news stories that "MRAM will soon replace flash." But as we have seen in news stories this year - some of those former "forever emerging" new nvms have emerged. They have emerging new business development sales channels.
  
  So the good news is we don't have to wait for graphene SSDs as there are already other silicon compatible persistent memory technologies which could enable ultra low power distributed industrial intelligence much sooner.
  
  In the past 10 years this applications area has been an alphabet soup of non volatile memories in search of applications to love them. This month - June 2017 - Crossbar published a slide deck which outlined such scenarios for its ReRAM.

• the use of much larger memories in IoT leveraging architecture and IP from phones and the cloud.
  
  This has been a long time in coming. And isn't here yet.
  
  I included this as a prediction for 2016. And 2 quarters later Marvell's FLC technology looked like an early match in that development stream. That was 1 of only 4 path lighting SSD technologies which I picked for a special mention at the end of 2016 which I had encountered across all SSD application markets - not just industrial.
  
  Don't expect too much too soon in tiered industrial memory, however. Even the better funded tiered memory market for the enterprise has been slower to ship products than expected due to a proliferation of competing solutions arising from the SCM DIMM wars market pulling in different directions and centered around multiple memory types.
  
  Some of the contenders in the enterprise nvm market may not even be suitable for industrial temperature operation yet (for example Micron's 3DX) whereas other nvms with high temperature remanence and possibly even good radiation resistance (such as Everspin's MRAM) haven't got the density (Gb per chip) to provide a complete solution and need to be supplemented by a tier of flash.

• discarding the conventional industrial market wisdom and exclusive reliance on the "solo SSD".
  
  Industrial SSDs have traditionally been used as solo devices. This saves space and power but places great dependence on the reliability of each single SSDs. And you know what that does to the price.
  
  The enterprise learned decades ago that incorporating array concepts for self repair and reliability enables systems reliability to be created from architecture and management and doesn't depend critically on the reliability - or cost - of any single drive.
  
  But hey - they've got big racks and big power supplies and engineers (or robots) who hang around day 24x7 sipping coffee (or electricity) itching to plug in new upgrades and replace faulty modules.
  
  That's in contrast to the typical industrial market when space and budgets are tight.
  
  For the industrial market I think the array direction is a story which will unfold and change the industry during the next 10 years.
  
  It's easier to see how this would work at the high wattage and high data processed value end:- simply as the evolution of the mobile datacenter.
  
  But at the low wattage end - in cars and IoT feeds - it will require the integration of processors and persistent memory at the SoC level and entirely new product architectures.
  
  It may be that the array level industrial SSD will always have to be a custom device rather than a standard COTS product - because weighting the internal design factors in software repairable industrial drive arrays will be very applications dependent and a slippery moving target compared to the stark simplicity that if everything depends on a single solo industrial SSD then it had better be good.

this may be a stupid question but...

have you thought of supporting a RAM disk emulation in your new "flash as RAM" solution?

• The results will reveal interesting performance anomalies (in the flash as DRAM emulation) and (if they exist) complex quality idiosyncrasies.
  
  That's because RAM disk software is mature and generally well behaved. But flash as a RAM software is relatively new. Running one as an emulation on the other means you can leverage well proven flash SSD benchmark tools to see if there are holes or spikes in the RAM emulation software.
  
  This isn't going to magic away the need for true RAM tier evaluation tools. Something I've touched on before in a news story is it realistic to talk about memory IOPS?
  
  But you will learn something useful from the gap between ideal behavior and what you measure.

• Another reason is this... In the present state of the market the need for pure play vanilla SSD storage (whether it's dressed up as AFA (capacity or performance optimized), or hybrid appliance or SDS is healthy and users will still be buying such quaint old fashioned products for many years to come. But....
  
  In the long term future... in another 5 to 10 years - such "storage focused" implementations of memory systems will become a smaller part of the overall local storage market. ("Local" being on your premises and not in the cloud. In the cloud we assume they will still find value based uses for cheap old slow stuff in the same way as they currently do with hard drives.)
  
  Instead - I think that "storage" itself may become just a software selectable setting in resource boxes which are really just big memory systems. The forerunners of which I hinted at in my blog - after AFAs - what's the next box?
  
  When all storage is made from memory chips - then the decision - sell it as memory or sell it as storage - may become a business and branding decision rather than a true representation of the technology in the box.