QuickLogic Announces Time Series EOS S3AI Development Platform for Endpoint AI and Machine Learning Applications

March 18 2019, 00:55
QuickLogic, the developer of ultra-low power multi-core voice-enabled SoCs, embedded FPGA IP, and Endpoint AI solutions, announced the launch of its EOS S3AI platform for endpoint Artificial Intelligence (AI) and Machine Learning (ML) applications. The new AI platform provides all the hardware blocks needed to enable low power AI at the connected device endpoint. It comes with time-series (continuous) sensors, including accelerometer / gyroscope / magnetometer motion sensors and two digital microphones (PDMs).

The new AI platform also includes seamless hooks to the SensiML Analytics Toolkit for sensor applications requiring time series (continuous) data analysis. This platform approach speeds system-level design and implementation, improves bandwidth, power, latency and security, while reducing many of the costly and time-consuming resources necessary to deploy AI-based IoT solutions. The EOS S3AI platform solution is ideal for endpoint applications, such as predictive maintenance, structural health monitoring, context awareness for wearables, and manufacturing process control.

This single chip solution utilizes the SensiML Analytics Toolkit to efficiently identify which complex feature extraction algorithms are ideal for the labeled data sets and generates the AI models and classifiers using machine learning techniques for code development. The AI model that is subsequently generated is uniquely optimized to fit in the EOS S3AI hardware blocks (eFPGA, uDSP and M4) and are programmed effortlessly by the SensiML Toolkit. This tight integration of hardware and software ensures that the AI algorithms created with the toolkit are optimized for the EOS S3AI platform, enabling low power and efficient endpoint processing with the help of hard accelerators and eFPGA feature extraction algorithms. 

The EOS S3AI platform allows the OEM to lower system and operating costs and improve reliability by eliminating the need for continuous high-bandwidth connectivity. Because the platform enables AI endpoint solutions to be developed easily and quickly with minimal data science and firmware engineering resources, endpoint application customers will benefit from lower product development costs while gaining valuable time-to-market advantages.

The EOS S3AI development platform and the SensiML Analytics Toolkit are available now.  

QuickLogic EOS S3
For hearables, wearables, smartphones, and other connected device applications with voice trigger and sensor management, QuickLogic is now shipping an ultra-low power SoC solution which combines always-on/always listening voice processing with fast, efficient and sophisticated sensor processing capabilities. The QuickLogic EOS S3 Voice and Sensor Processing solution supports 2- to 4-mic beamforming and noise-suppression with dedicated uDSP (Flexible Fusion Engine) for sensor processing, On/Off-body detection for lowest power mode, with a low-voltage (0.85 VDD) option for 33% power reduction.

The ultra-low power EOS S3 enables OEMs to extend battery life while designing in sophisticated, always-on sensing capabilities on mobile devices. Advanced sensor algorithms such as voice triggering, motion compensated heart rate monitoring, and indoor navigation can be achieved at significant power reduction compared to competing MCU-based solutions.

Unlike traditional MCU-based solutions, the EOS S3 is a multi-core, sensor processing system that enables sophisticated algorithm partitioning to facilitate the lowest possible power consumption for the designated task. The EOS S3 employs not only fundamental, but also very sophisticated, always-on, context-aware sensing capabilities while staying well within the strict power budgets of smartphone, wearable, and hearable designs.

The 10MHz DSP-like processor supports always-on computational processing at one forth the power, while the eFPGA enables custom logic functions and I/O expansions. For voice processing, the EOS S3 solution features a hard-coded Low Power Sound DEtector (LPSD) and PDM to PCM conversion that minimizes audio processing power, with an ARM Cortex-M4 with FPU, up to 80MHz and 512 KB SRAM for general purpose processing and running O/S, also including SPI Master/Slave I2C, UART, RTC Oscillators, ADC, and LDO.

In China, the EOS S3 solution was successfully implemented in JingDong Group's (JD.Com) latest U-LIFE N1 neck-worn always-on wake word Bluetooth enabled Smart Earphone Headset. The new U-LIFE N1 headset features active noise cancelation and is targeted at consumers with an active lifestyle. It features a streamlined neck-worn design with an intelligent voice assistant which has a voice recognition rate of 95%. The reliability and the accuracy of the headphone enables hands-free operation, allowing users to easily switch between listening to music, taking a phone call using only their voice and providing a functional experience. The embedded QuickLogic EOS S3 device listens for the wake word, and then activates the rest of the device only when necessary - saving power and extending battery life.
related items