NDP250

Our family of Neural Decision Processors (NDPs) are specifically designed to run deep learning models, providing 100x the efficiency and 10/30x the throughput of existing low-power MCUs.

Syntiant leads in edge-AI with custom silicon enabling deep learning on battery-powered devices. Their technology supports real-time data processing and decision-making with minimal power consumption, from hearing aids to automotive applications. Syntiant’s solutions empower developers to quickly deploy efficient deep learning models across a wide range of edge devices, revolutionizing applications in audio, video, and sensor processing.

Characteristics

Additional information

DNN Core Memory	6MB
DNN Core Clock	Same as DSP
DNN Core Features	All Networks, Fixed Point
MCU Memory	0.5MB
MCU Clock	Same as DSP
DSP Memory	1.5MB
DSP Clock	0,8V: < 80MHz, 0,9V: < 120MHz
DSP Features	HiFi3: Floating-point
Package [mm]	6,1mm x 5,1mm 120-ball eWLB
Product status	Sampling

The Syntiant® NDP250 Neural Decision Processor™ is a special-purpose chip for edge-AI processing in battery powered devices and other power constrained systems. Target applications include vision, audio, Automatic Speech Recognition (ASR), and Text-To-Speech (TTS) processing. Built using the Syntiant Core 3™ programmable deep learning architecture, NDP250 is designed to natively run multiple Deep Neural Networks (DNN) on a variety of architectures. The NDP250 delivers a 5x increase in ML performance compared to the previous Syntiant Core 2™ NDP120 and NDP200 devices. A programmable Hifi3 DSP is available for feature extraction and signal processing. The integrated power management (PMU) allows single power rail operation. Combined with the integrated PLL this provides further systems cost reduction opportunities.

Broad support of concurrent neural network: fully connected, 1D, 2D & Depth-wise convolution, average and max pooling, recurrent including LSTM and GRU, attention and many other layers
Up to 6M neural parameters in 8-bit mode
2 x 11-wire direct image (DVP) interfaces
3 x Stereo PDM digital microphone/I2S/TDM input interfaces
1 x I2S/TDM PCM output interface
1 x Quad-SPI and 2 xI2C controller and target for multi-modal sensor fusion
1 x UART and 2 x PWM interfaces
1 x 5-wire JTAG debug interface
58 GPIO pins
Programmable HiFi 3 DSP with 1.5MB of SRAM and Syntiant custom instructions
Embedded Arm Cortex-M0 with 512KB of SRAM
Up to 120MHz internal operating frequency
Input holding tanks to support pre-roll with up to 10 seconds of 550×700 compressed Motion-JPEG image recording at 5 FPS
Low power PLL for flexible clock input
Integrated low power oscillators for reference clock
1.8/3.3V supply with embedded Power Management Unit (PMU)
Onboard firmware security and authentication
Software Development Kit (SDK) integrates in any software environment
Training Development Kit (TDK) to enable the use of standard frameworks such as TensorFlow for customer-programmed applications
6.1mm x 5.1mm 120-ball eWLB package (0.5mm pitch)

The NDP250 supports dozens of application-defined imaging, speech, audio, and sensor behaviors for a variety of use cases including:
Person presence detection with pre-roll
Object classification
On-device ASR
Motion tracking
Acoustic event and scene classification
Multi-microphone full duplex voice communication
Multi-sensor fusion

NDP250

Our family of Neural Decision Processors (NDPs) are specifically designed to run deep learning models, providing 100x the efficiency and 10/30x the throughput of existing low-power MCUs.

Additional information

DNN Core Memory	6MB
DNN Core Clock	Same as DSP
DNN Core Features	All Networks, Fixed Point
MCU Memory	0.5MB
MCU Clock	Same as DSP
DSP Memory	1.5MB
DSP Clock	0,8V: < 80MHz, 0,9V: < 120MHz
DSP Features	HiFi3: Floating-point
Package [mm]	6,1mm x 5,1mm 120-ball eWLB
Product status	Sampling

Broad support of concurrent neural network: fully connected, 1D, 2D & Depth-wise convolution, average and max pooling, recurrent including LSTM and GRU, attention and many other layers
Up to 6M neural parameters in 8-bit mode
2 x 11-wire direct image (DVP) interfaces
3 x Stereo PDM digital microphone/I2S/TDM input interfaces
1 x I2S/TDM PCM output interface
1 x Quad-SPI and 2 xI2C controller and target for multi-modal sensor fusion
1 x UART and 2 x PWM interfaces
1 x 5-wire JTAG debug interface
58 GPIO pins
Programmable HiFi 3 DSP with 1.5MB of SRAM and Syntiant custom instructions
Embedded Arm Cortex-M0 with 512KB of SRAM
Up to 120MHz internal operating frequency
Input holding tanks to support pre-roll with up to 10 seconds of 550×700 compressed Motion-JPEG image recording at 5 FPS
Low power PLL for flexible clock input
Integrated low power oscillators for reference clock
1.8/3.3V supply with embedded Power Management Unit (PMU)
Onboard firmware security and authentication
Software Development Kit (SDK) integrates in any software environment
Training Development Kit (TDK) to enable the use of standard frameworks such as TensorFlow for customer-programmed applications
6.1mm x 5.1mm 120-ball eWLB package (0.5mm pitch)

The NDP250 supports dozens of application-defined imaging, speech, audio, and sensor behaviors for a variety of use cases including:
Person presence detection with pre-roll
Object classification
On-device ASR
Motion tracking
Acoustic event and scene classification
Multi-microphone full duplex voice communication
Multi-sensor fusion

Characteristics

Additional information

DNN Core Memory	6MB
DNN Core Clock	Same as DSP
DNN Core Features	All Networks, Fixed Point
MCU Memory	0.5MB
MCU Clock	Same as DSP
DSP Memory	1.5MB
DSP Clock	0,8V: < 80MHz, 0,9V: < 120MHz
DSP Features	HiFi3: Floating-point
Package [mm]	6,1mm x 5,1mm 120-ball eWLB
Product status	Sampling

NDP250

Our family of Neural Decision Processors (NDPs) are specifically designed to run deep learning models, providing 100x the efficiency and 10/30x the throughput of existing low-power MCUs.

Broad support of concurrent neural network: fully connected, 1D, 2D & Depth-wise convolution, average and max pooling, recurrent including LSTM and GRU, attention and many other layers
Up to 6M neural parameters in 8-bit mode
2 x 11-wire direct image (DVP) interfaces
3 x Stereo PDM digital microphone/I2S/TDM input interfaces
1 x I2S/TDM PCM output interface
1 x Quad-SPI and 2 xI2C controller and target for multi-modal sensor fusion
1 x UART and 2 x PWM interfaces
1 x 5-wire JTAG debug interface
58 GPIO pins
Programmable HiFi 3 DSP with 1.5MB of SRAM and Syntiant custom instructions
Embedded Arm Cortex-M0 with 512KB of SRAM
Up to 120MHz internal operating frequency
Input holding tanks to support pre-roll with up to 10 seconds of 550×700 compressed Motion-JPEG image recording at 5 FPS
Low power PLL for flexible clock input
Integrated low power oscillators for reference clock
1.8/3.3V supply with embedded Power Management Unit (PMU)
Onboard firmware security and authentication
Software Development Kit (SDK) integrates in any software environment
Training Development Kit (TDK) to enable the use of standard frameworks such as TensorFlow for customer-programmed applications
6.1mm x 5.1mm 120-ball eWLB package (0.5mm pitch)

The NDP250 supports dozens of application-defined imaging, speech, audio, and sensor behaviors for a variety of use cases including:
Person presence detection with pre-roll
Object classification
On-device ASR
Motion tracking
Acoustic event and scene classification
Multi-microphone full duplex voice communication
Multi-sensor fusion

NDP250

Characteristics

Additional information

NDP250

Additional information

Characteristics

Additional information

NDP250

Products

Services

Legal

Information about this product

NDP250

Samples request