Thursday, February 14, 2019

More about AIStorm

Venturebeat, ElectronicsWeekly, EETimes report more details on AIStorm AI-on-Sensor startup:

  • AIStorm has been founded in 2011 and has been in stealth mode till the recent announcement of $13.2M round A financing.
  • AIStorm’s patented chip design is capable of 2.5 Tera Ops and 10 Tera Ops per watt, which said to be 5x to 10x lower than the average GPU-based system’s power.
  • The company uses a technique called switched charge processing, which allows the chip to control the movement of electrons between storage elements
  • "The TowerJazz pixel is part of our input layer, so the charge comes from sensors, they produce electrons, and we multiply and move them" - says AIStorm
  • AIStorm tested its first chip this month and plans to ship production orders next year
  • The company’s first products are to be made in 65nm or 180nm process
  • AIStorm is planning Series B round for follow up products in 28-nm and possibly finer nodes
  • The production chips are aimed to be compatible with popular AI frameworks such as TensorFlow


  1. the precision of analog computation is proportional to sqrt(area) and that of digital counterpart is rather 2^area. when the scaling-down continues, the analog computation becomes efficient only in some very specific applications. Hope that NN could be this specific application but with large market scale !

  2. The suggested rule of thumb is not applicable in the case of AIStorm's technology. The solution is scalable, and is not relegated to a limited set of NN applications but is widely applicable. The solution really is a revolutionary breakthrough and completely different to any other AI solution out there.

    1. Any technical publications you could cite to back up these bold claims on "revolutionary"?


All comments are moderated to avoid spam and personal attacks.