Embedded Edge AI Machine Learning on a Low Cost Chip (LSQML000)

Embedded Design Developed Faster Enhanced Revenue Competitive Edge

Turn weeks of development testing into days
Auto-determine production/operations tolerances and maximise yield
Auto fault/wear monitoring to optimise service revenue and reliability
Competitive edge: IP enhancements discovered via exhaustive and automatic data analysis

Weeks of development become days

Hours/days/weeks sometimes can be spent doing laborious tests to determine look-up table values so that a control system can be effected. If there’s more than one independent variable, say temperature, pressure, voltage, it can be onerous to design and do the testing necessary. The LSQML000 does it all automatically. Just bring the signals (up to 6) in on the ADCs or SPI/I2C if digital, and the chip will generate the look tables automatically when real world values are encountered. It saves a great deal of time.

Optimum Operations Yield

Learning doesn’t just deliver look up tables and transfer functions automatically, it allows for the capture of the natural or real tolerances of the system. SPC (statistical process control) is integral to the LSQML000. Furthermore, the tolerancing is dynamic (tolerances for each setting). Gone are the days of relatively loose tolerance setting leading to either restricted yield or a failure to discriminate non-conforming products.

Optimised service revenue and monitoring

Enhanced revenue from planned servicing that is tuned to minimise risk of unpredicted failure is in the gift of the LSQML000. Changes in system performance with use and over time are monitored and corrected. Furthermore, changes with time are subjected to robust statistics, such as SPC, and failure prediction results. These factors lead to avoidance of unplanned failure and revenue from intelligently scheduled service.

Milking IP opportunities with embedded AI (LSQML000) microchip

Traditional embedded design is largely based on binary thinking. Is the pressure greater than ?, therefore turn pump on. The problems with this is that there is considerable redundancy in the data collected. What else could be learned from the complete pressure data set, camera imaging, current, etc. Humans typically don’t have the time to pore over this, but a chip like the LSQML000 can do it automatically. It leads to knowledge that can reveal intellectual property that in turn gives a client’s system competitive advantage. There is a revolution taking place in embedded design that finds its origins in machine learning and AI – companies that overlook this risk being left behind.

Embedded Design and AI: Where is it going?

Using AI in embedded design requires a shift in design paradigm. Traditionally, a design is initiated on the basis of it being feasible or not. If there is doubt there will be feasibility work required before a project is initiated. The addition that is not usually part of this method that AI brings in is that what is needed may not be known until some AI data learning and ground work has been done. The LSQML000 allows this to be done without onerous delay or cost.