Always-on-sensing: don’t waste power in digits

Always-on-sensing: don’t waste power in digits

Technology News |
By eeNews Europe

Founded in 2015, the West Virginia startup says its RAMP “analyze-first” edge-architecture drastically minimizes data handling, cutting the volume of data handled by up to 100x for always-on applications while maximizing battery life thanks to a 10x power reduction.

eeNews Europe caught up with Tom Doyle, Aspinity’s CEO and founder to learn more about the company’s unique technology.

“Today, a lot of processing goes from the cloud back to the edge, but it is still challenging to get the required performance, accuracy and long battery life” explained Doyle, referring to battery-operated always-listening smart home security devices, voice-first smart speakers and wearables, and industrial vibration monitoring systems. The reason, he argues, is that these systems are digitizing too much irrelevant data. So the idea is to leverage an analog neural network onboard an analog processor to sort out and identify the data that would deserve to be digitized (in a simplified format) for subsequent use by an application processor. This way, the application processor and other components can stay off most of the time.

Aspinity’s analyze-first system architecture.

“We have a purely analog processor, all of the sensor data classification is done within the analog domain” the CEO emphasizes, noting that his company has developed libraries of AI algorithms based on training data, which it can then load into its RAMP chip. This means the same chip can be updated precisely for different voice detection patterns, using the same hardware.

“We’ve demonstrated system wake up for specific alarms, where the RAMP does not wake the system if other alarms are detected. We can discriminate, for example we’ll detect voice but we won’t detect door knocks”, Doyle continued.

Putting some figures on current and power consumption, the CEO said always-on sound processing using the RAMP would draw from 10 to 15uA, whereas a “digitize-first” architecture built around an analog-to-digital converter and a digital signal processor would not be out of the mA range.

Key power savings come from the fact that the RAMP platform is able to detect and classify events from background noise in the raw analog sensor data before the data is digitized. This approach eliminates the higher-power processing and transmission of irrelevant data.

Aspinity’s field-programmable
analog array (FPAA).

Aspinity does not want to share much about its patented RAMP technology, except that it leverages the nonlinear characteristics of a small number of transistors, incorporating modular, parallel and continuously operating analog blocks as part of a neural network. As reported in a 2015 IEEE paper “A Low-Power Field-Programmable Analog Array for Wireless Sensing”, the chip essentially consists in a field-programmable analog array (FPAA) sporting a 8×10 array of computational blocks. Hence it supports 8 identical channels and 10 function-specific stages. The company hope to have a demonstrator for release next year.

“We have silicon we can sample today. The RAMP is scalable and now we only use between 10 and 20% of its resources” Doyle told eeNews Europe. “When we get to the point we understand the full requirements of our customers, we’ll be able to scale it up or down.

The idea is to sell a silicon line that takes multiple inputs from accelerometers, cameras, microphones, but the company does not discard the opportunity to see its device integrated into microphones.

“We envision our RAMP to be programmable by everyone, we have a software development kit and we’ll offer a programming environment with libraries. Users won’t have to be analog experts” said Doyle.

“For industrial use like preventive maintenance, sensor experts look for spectral components, a peak in frequency or magnitude, they’ll be able to detect that peak and track it. They can also push their own training data into our RAMP model and then use our compiler to compile it into our RAMP chip through an SPI port, via a microcontroller” continued the CEO.

Reconfigurable analog blocks make the RAMP chip very versatile.

The RAMP platform’s analog blocks can be reprogrammed with application-specific algorithms to analyze raw analog data from multiple types of sensors, such as accelerometers used for industrial vibration monitoring. Instead of a predictive maintenance system that continuously digitizes thousands of points of data to monitor the trends in the changes of certain spectral peaks, RAMP can sample and select only the most important data points, compressing the quantity of vibration data by 100x and dramatically decreasing the amount of data collected and transmitted for analysis.

Today, Aspinity’s initial focus is on the detection of voice and acoustic events for the fast-paced consumer market such as voice-activated home assistants and smart home monitoring. Indeed, according to Juniper Research, the installed base of digital voice assistants will triple to 8 billion by 2023, and always-on voice-first devices, such as smart speakers and wearables/hearables are among the largest and fastest-growing market segments.

Next, the startup wants to address the industrial market and within the next couple of years, it aims to work on solutions for the biomedical market, to detect heart rate problems with a wearable device that could last one or two years on a single battery.

Aspinity –

Related articles:

Audio rises for event detection

Embedded sound recognition as a trigger for smart appliances

If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News


Linked Articles