Blumind

What Blumind does

Blumind is a Canadian deep-tech startup bringing machine-learning inferencing to the Far Edge, placing it on devices and sensors in all environments. Their breakthrough all-analog, in-memory inferencing engine slashes power consumption, latency, and silicon area by orders of magnitude versus digital approaches. Blumind's proprietary AMPL™ core architecture is at the heart of all their products, designed to mimic the human brain as closely as possible by creating an all-analog AI neural network architecture with the most efficient power profile possible, using standard, cost-effective CMOS process technology.

Products

• DevicesAdvanced all-analog devices offering cutting-edge solutions for audio, time series data, and visionary visual applications.
• BM110All-analog device for always-on audio and time series data applications.
• BM210All-analog device for always-on video and image classification applications.
• AMPL Intellectual PropertyBlumind's proprietary AMPL intellectual property, an all-analog neural signal processor IP for edge AI applications, designed on standard advanced node CMOS technology for integration into high volume SoC and MCU products.
• Blumind ChipletNeural Signal Processor known-good-die for system in package integration.
• SoftwareNeural system processor inference products that use industry standard software flows including PyTorch and TensorFlow.
• Blumind AI integration suiteA translator that maps coefficients after quantization and compression into a weights file for the Blumind device, allowing networks to be retrained and updated with new weights.

Key capabilities

• All-analog compute
• Lowest system power
• Ultra-low latency
• Low cost solutions
• Standard cost-effective CMOS process technology
• AMPL™ technology
• Easy-to-use with industry standard software tools
• Minimizes PVT and drift effects
• High accuracy and scale
• All-analog neural signal processor technology
• Combines machine learning, precision analog signal processing, and brain-inspired computing
• Deterministic and precise inferencing performance
• Longest battery life for always-on applications
• First all-analog AI on advanced standard CMOS architected to fundamentally mitigate process, voltage, temperature and drift variations
• Direct Analog Sensor Input
• No ADC & DAC in the neural network core
• No SRAM
• No High Speed Clock
• No Added Processing Steps
• No Novel Process Technology Modules
• Supports Sustainable Solutions
• Patented architecture does not use voltage or current steering
• Combination of ratio-metric circuits and easily regulated references produce deterministic results
• Lowest Total System Cost
• PVT and Drift Mitigation
• Works with analog (or digital) sensors

Use cases

• Machine-learning inferencing on devices and sensors
• Always on Key Word Detection
• Voice UI
• Voice control/commands
• Environmental classification
• Biometric pattern recognition
• Smart industrial sensors
• Real-time AI inferencing
• Always on Visual Wake Word
• Visual wake word/trigger
• Gesture identification
• Biometric Identification
• Defect identification/classification
• Threat/hazard identification
• Visual inspection/QA
• Real-time localized image classification
• Real-time data processing for wearables
• Always-on voice commands for connected home
• Responsive visual wake trigger for connected home
• Intelligent automation for connected home
• Rapid time series data analysis for industrial, agriculture & medical
• Precise image classification for industrial, agriculture & medical
• Cognitive-like computing for industrial, agriculture & medical
• Real-time surveillance for safety & security
• Facial recognition for safety & security
• Predictive analysis for safety & security
• EV battery life management
• Autonomous vehicles
• Real-time diagnostics for smart mobility
• Drone collision avoidance
• Intelligent automotive systems
• Collision Avoidance in UAV
• Image Analysis for cameras
• Target identification for Radar & Sonar
• LiDAR imaging
• Night Vision Enhancement
• Voice Interface for smart glasses
• Adding trigger keyword detection to wearable electronics
• Environment classification for True Wireless Stereo headphones
• Speech-to-intent (Voice UI) for True Wireless Stereo headphones
• Adding always-on key word detection, environmental classification and speech to intent (Voice UI) all-analog AI to smart glasses
• Localized always-on voice interaction for smoke detectors, remote controls, security systems and white goods
• Real-time edge AI analysis of visual or time-variant data for industrial, agriculture and medical products
• Direct analog sensor data analysis from vibration, acoustic, spectroscopy, EKG, moisture, pH, pressure or temperature sensors
• Localized visual inspection solutions without cloud connectivity
• Real-time monitoring for medical devices
• Process Control & Excursion Management
• Soil Moisture/Irrigation
• Visual Produce Inspection
• Food Safety & Cold Storage
• Animal Management
• Adding voice or vision to mobile battery-based products
• Gesture and natural language interfaces for wearables and connected home
• Voice and gesture control for home automation
• Making appliances more efficient and enhancing user experience
• Front door AI deployment

AI approach

Blumind develops all-analog AI neural network architectures and hardware solutions (devices, chiplets, IP) for edge AI inferencing. Their proprietary AMPL™ technology uses standard CMOS processes to deliver ultra-low power, low latency, and low-cost AI compute, specifically mitigating PVT and drift effects inherent in analog circuits. They focus on 'always-on' applications for audio, video, and time series data, and their software integrates with industry-standard frameworks like PyTorch and TensorFlow for model training and deployment to their hardware.

Tech named: all-analog AI neural network architecture, AMPL™ technology, CMOS process technology, in-memory inferencing engine, PyTorch, TensorFlow, neural signal processor, ratio-metric circuits, analog signal processing, brain-inspired computing

Industries served

• Semiconductor Manufacturing
• IoT
• Automotive
• Smart Home
• Smart City
• Wearable & Personal Devices
• Industrial
• Agriculture
• Medical
• Connected Home
• Safety & Security
• Smart Mobility
• Mil Aero
• Consumer Electronics
• Healthcare

What it says sets it apart

• All-analog compute offering the lowest system power, latency and cost solutions
• Breakthrough all-analog, in-memory inferencing engine
• Slashes power consumption, latency and silicon area by orders of magnitude versus digital approaches
• AMPL™ technology delivers industry standard inferencing performance but with up to x1000 lower power
• Ultra-low latency compared to legacy digital approaches
• No-compromise easy to deploy solutions using industry standard software tools
• Uses standard CMOS processes
• Architecture minimizes PVT and drift effects
• Delivers accuracy and scale needed for today's leading edge use cases
• All-analog AI neural network architecture delivers the most efficient power without compromise
• Proprietary architecture does not use voltage or current steering and does not use ADCs or DACs in the neural network core
• Combination of ratio-metric circuits and easily regulated references produce deterministic results
• Lowest total system cost for industrial, agriculture and medical products
• AI solutions are up to 2-orders of magnitude lower power than alternatives
• Smallest edge AI footprints enable retrofitting to existing form factors without mechanical redesign
• Ultra-low latency for real-time classification of video, audio and time series data
• Works with analog (or digital) sensors for power, area and cost efficiency
• Localized always-on edge AI without cloud connectivity
• AMPL™ core architecture is the lowest power commercially available neural networks solution on a given process node

Funding rounds we track

From the AI funding tracker — rounds as reported by the linked publications.