Lattice FPGA technology stack enables AI in edge devices
Since the company’s FPGAs are widely used as highly configurable I/Os for bridging chips in low power edge devices, it would only take a new code compile to smarten up edge devices, hardly a redesign, claims the company, virtually offering AI for a few cents.
“One of the first things that customers tell us is, we like AI, we understand the value but we are not willing to do significant changes to our design to fit AI. We need to use existing sensors and processors. So AI chips should have very flexible I/Os, definitely not something ASICs are good at”, Deepak Boppana, senior director, product and segment marketing at Lattice Semiconductor told eeNews Europe.
The Lattice sensAI stack is optimized for ultra-low power consumption (under 1mW to 1W) in packages ranging from 5.5 to 100mm2 and boasts excellent interface flexibility (MIPI CSI-2, LVDS, GigE, etc.), facilitating edge computing close to the source of data.
Lattice’s sensAI stack includes modular hardware platforms, first the ECP5 device-based Video Interface Platform (VIP), including the Embedded Vision Development Kit, and the iCE40 UltraPlus device-based Mobile Development Platform (MDP). It also comes with two new IP cores, a Convolutional Neural Network (CNN) accelerator optimized for the ECP5 FPGA with support for variable quantization (16/8/1) for weights and activation and a Binarized Neural Network (BNN) accelerator optimized for the iCE40 UltraPlusFPGA with support for 1 bit weight, 1 bit activation quantization.
Lattice also packaged software tools including a neural network compiler tool for Caffe/TensorFlow to FPGA, allowing designers to implement networks developed using standard frameworks into Lattice FPGAs without prior RTL experience.
Lattice Radiant design software and Lattice Diamond design software complete the sensAI stack together with several reference designs for face detection, key phrase detection, object counting, face tracking, and speed sign detection. Lattice bundles this with an eco-system of design service partners able to deliver custom solutions for broad market applications, including smart home, smart city, and smart factory.
In the future, third party AI IP cores may be included in the stack. Boppana envisions that the low-power FPGAs could be used to perform always-on inferencing and key phrase detection or face/object detection for smart TVs or smart vending machines that would only turn on when they detect a face.
Lattice Semiconductor – www.latticesemi.com/sensAI
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
