Ai Linear’s Mission:

Enable distributed intelligence (ultra-edge computing) by offering mixed-signal (analog and digital) building-blocks and solutions for low-power, low-cost, small-size, low-latency, high-Safety/privacy, and Application Specific Standard ICs (ASICs) that facilitate making machine learning (ML) inference ICs in/near sensors/devices 

Ai Linear's Competitive Advantage:

  • Intellectual Property: Founder of the company has a growing portfolio of 35+ U.S. Patents (including pending) in ultra-low-power analog/mixed-signal solutions for machine learning and sensor interface
  • Team: Ai Linear team has decades of Digital, Analog, and Software R&D and operations experience (in Silicon Valley, California USA) with a proven track record of making successful, profitable Integrated Circuit products
  • Focus: Dedicated & Configurable Low-Power Machine-Learning  Accelerators near/on sensors for Industrial & Medical Markets

Ai Linear's Vision of the future of Distributed Intelligence

Ai will be Everywhere and it will be Always-On (Ai on Sensors: Ai Dust).  “Ai Everywhere” neither needs to rely on the INTERNET nor on the CLOUD for intelligence:

Heath, Safety, and Efficiency:  Distributed Intelligence (i.e., Ultra-edge machine learning, Ai near/in sensors and devices, smart sensors) will be everywhere to quietly help keep people, the environment, industries, and things safe, healthy, and efficient with unimaginable new capabilities.  Attractiveness of privacy and safety aspects of Ai near sensors with pre-defined and limited intelligence will accelerate its adoption in the face of growing concerns and mistrust of over-reaching, hard to regulate, and increasing scope of intelligence of the CLOUD

Ultra-portability:  Ai in sensors and devices will be fueled by (1) applications needing continuous and “always-on” monitoring that require ultra-low-power battery-less portability in light of limited access to a power source, (2) application that are hard to reach or maintain and are size-constraints; (3) objectives to detect “important” outputs from sensors and to avoid sending “unimportant” sensor outputs to the CLOUD which wastes power and is costly for its dependence on communications (I/O) with the CLOUD for intelligence; (4) applications that cannot afford “latency” associated with communicating with the CLOUD (5) availability of custom ultra-low-power limited-data I/O that may not necessarily be Wi-Fi/Bluetooth/5G, and may be more simple like I/O used in car-key, baby-monitor, or garage-opener

Less Wasteful Data on Cloud, New Data/Analytics Software:  Distributed Intelligence will pre-filter wasteful data right near sensors before they are sent to the CLOUD, that would give birth to new Software and Data Analytics derived from Ai in sensor and devices, which will further drive the fusion of sensors with machine learning capability

Machine Learning + Sensors = Ai in Sensors and Devices:  We are at the tip of the iceberg for a WAVE of new data to be generated by smart sensors and their cross-pollination such as images, sounds, vibration, leaks, location, smells, tastes, pressure, toxicity, temperature, humidity, gases, acceleration, shocks, sweat, coughing, falling, etc.

Shift from Cloud to Fog:  There will be a major shift of intelligence away from the CLOUD to the FOG because of (1) PRIVACY & SAFETY for some customers or applications that cannot trust the CLOUD with their information wherein, they will want to keep it “on premises”. So, the FOG is in effect a “private-cloud”; (2) LATENCY where in general there are no run time guarantees in cloud processing. By employing a FOG or a private-cloud, a customer retains control of such private-cloud and can guarantee availability as required by their application. (3) Application specific ASIC Accelerators will become available that will be tailored to application specific data-loads, wherein such FOG based platforms accommodate significantly less power-hungry communication ports with application specific Ai in/near sensors and devices

Hybrid computing will gain market acceptance: Analog and mixed-mode accelerator ICs (utilizing novel analog architectures and algorithm) will be paired-up with digital processor ICs (utilizing application specific digital architectures and algorithms) for segmented signal processing, wherein the digital processor will perform precise computations after the analog accelerators performs approximate computation on incoming data, for substantial overall system cost and power consumption savings

For the FOG, Growth in Application Specific AI accelerators with thin-software will surpass that of General Purpose and Fully-Programmable Computation engines:  Start-ups chasing AI Software based on general-purpose computational platforms in the FOG will continue to face an up-hill battle on two fronts: (1) innovations in software will less likely receive patent protection because the USPTO remains reluctant to issue software-based patents; (2) Large conglomerates will dominate the FOG-based AI software space and FOG-based general-purpose computation engines, leaving fewer low-hanging fruits for start-ups aiming for FOG-based Ai software space; (3)  Also, large conglomerates will push for low-power low-cost solutions that are application specific Ai accelerators with thin software for the FOG, as opposed to general purpose fully software programmable computation engines for the FOG

Analog is a Natural Fit for Ai in/near Sensors: The real world is analog and not digital.  The analog world can interface more naturally with analog ICs.  Real world signals are more approximate, and analog solutions can perform approximate computation with less power and less cost than digital solutions.   Additionally, because silicon manufacturing is approaching the end of Moore’s Law, going forward OEMs can no longer bet on adopting expensive digital IC solutions now based on a hope that digital IC densities and costs can be cut in half every 18 months (as they have done for the past 4 decades).  As such, it may be a risky bet to count on prices and power consumption of programmable/fully digital chips to drop low enough for ultra-portable portable and size constrained Distributed Intelligence applications, considering that manufacturing of ‘bleeding edge’ ICs will remain dominant in the hands of a few heavy-weight fabrication suppliers with little motivation to substantially lower prices from here

Innovations in Mixed-Signal R&D Tools and Fabrication: The end of Moore’s law (that helped make SMALLER and more DENSE and CHEAPER ICs and transistors) does not mean the end of device physics’ innovation to make transistors more ACCURATE.  It is not a matter of if, but when silicon fabrication scientists will substantially enhance transistor and device precision. This will empower and enable Mixed-Signal IC development flow to approach that of pure digital IC development flow. CAD/Automation/computation tools will substantially improve Time-To-Market and significantly reduce the NRE of Mixed-Signal ICs by offering ease of abstraction, simulation, and synthesis of Mixed-Signal designs (comparable to the digital IC flow), which will further accelerate practicability and the adoption of Distributed Intelligence based in analog and mixed-signal ICs

As Hybrid Computing and Analog Computers will rise, Analog Software will be born:  With the ends of Moore’s law in sight, innovation in silicon manufacturing shifts from “more DENSE transistors” to “more ACCURATE transistors” which will help the making of rugged analog accelerators and then the adoption of hybrid computing (analog accelerators assisting digital accelerators). Consequently, new and efficient and standalone analog computers become a reality.  Consequently, new categories of thin and efficient software and tools tailored to analog and mixed-signal computation engines will rise