What you’ll learn:
- What’s new at BrainChip?
- The implementation of SNNs in some of the company’s latest chips and systems.
Spiking neural networks (SNNs) are a form of neuromorphic computing that provides a power- and computational-efficient artificial-intelligence (AI) model implementation. In the video (view above), BrainChip’s Chief Marketing Officer, Steven Brightfield, highlights some of the chips and systems mentioned here, including the new AKD1500. The earlier AKD1000 is available as a chip and in a number of form factors, such as M.2 and a PCI Express (PCIe) card (Fig. 1).
The AKD1500 can deliver 800 effective GOPS using only 1 mW/GOPS. BrainChip’s software suite can convert AI models from platforms like TensorFlow and PyTorch, so they work with this SNN AI accelerator to provide a faster and more power-efficient hardware implementation. The AKD1500 has an SPI and a PCIe interface for host support.
The chip and M.2 versions are useful, but BrainChip also released the AkidaTag reference design (Fig. 3). The reference design is a combination of the company’s AKD1500 AI accelerator with Nordic Semiconductor’s nRF5340 system-on-chip. The latter provides host processing support and Bluetooth connectivity. A matching smartphone app provides wireless feedback and control.
The AkidaTag uses its SPI interfaces to support sensors and peripherals. It can also interface with external serial-memory chips. The AkidaTag and other BrainChip platforms work with the Edge Impulse and BrainChip metaTF software, which is supported by the Akida Model Library.
More Content on Spiking Neural Networks
Podcast: For Ultra-Low-Power Neural Networks, Try Spiking Neural Networks
About the Author
William G. Wong
Senior Content Director - Electronic Design and Microwaves & RF
I am Editor of Electronic Design focusing on embedded, software, and systems. As Senior Content Director, I also manage Microwaves & RF and I work with a great team of editors to provide engineers, programmers, developers and technical managers with interesting and useful articles and videos on a regular basis. Check out our free newsletters to see the latest content.
You can send press releases for new products for possible coverage on the website. I am also interested in receiving contributed articles for publishing on our website. Use our template and send to me along with a signed release form.
Check out my blog, AltEmbedded on Electronic Design, as well as his latest articles on this site that are listed below.
You can visit my social media via these links:
- AltEmbedded on Electronic Design
- Bill Wong on Facebook
- @AltEmbedded on Twitter
- Bill Wong on LinkedIn
I earned a Bachelor of Electrical Engineering at the Georgia Institute of Technology and a Masters in Computer Science from Rutgers University. I still do a bit of programming using everything from C and C++ to Rust and Ada/SPARK. I do a bit of PHP programming for Drupal websites. I have posted a few Drupal modules.
I still get a hand on software and electronic hardware. Some of this can be found on our Kit Close-Up video series. You can also see me on many of our TechXchange Talk videos. I am interested in a range of projects from robotics to artificial intelligence.
Facts Only
BrainChip produces Spiking Neural Networks (SNNs) for neuromorphic computing.
The AKD1500 chip delivers 800 effective GOPS.
The AKD1500 consumes 1 mW/GOPS.
The AKD1500 features SPI and PCIe interfaces.
The AKD1000 is available as a chip, M.2 module, and PCIe card.
BrainChip software converts AI models from TensorFlow and PyTorch.
The AkidaTag is a reference design combining the AKD1500 with Nordic Semiconductor’s nRF5340 system-on-chip.
The nRF5340 provides host processing and Bluetooth connectivity.
The AkidaTag includes a matching smartphone app for wireless control.
The AkidaTag supports sensors, peripherals, and external serial-memory chips via SPI.
The platforms utilize Edge Impulse, BrainChip metaTF software, and the Akida Model Library.
Executive Summary
BrainChip is advancing neuromorphic computing through the implementation of Spiking Neural Networks (SNNs), designed to offer higher power and computational efficiency compared to traditional AI models. The current hardware lineup includes the AKD1000 in various form factors and the newer AKD1500, which achieves 800 effective GOPS at a power efficiency of 1 mW/GOPS. To ease adoption, BrainChip provides a software suite capable of converting existing models from TensorFlow and PyTorch into SNN-compatible formats.
Beyond standalone chips, the company has introduced the AkidaTag reference design. This system integrates the AKD1500 AI accelerator with a Nordic Semiconductor nRF5340 SoC, enabling Bluetooth connectivity and host processing, managed via a dedicated smartphone application. The ecosystem is further supported by integration with Edge Impulse and the metaTF software library, facilitating the connection of the hardware to various sensors and peripherals.
Full Take
The narrative presented is a classic technology showcase, positioning SNNs as the inevitable evolution for power-constrained AI. The strongest version of this claim is that moving from traditional deep learning to neuromorphic "spiking" architectures allows for a radical reduction in energy consumption—essentially mimicking the efficiency of the human brain to enable "edge" intelligence where batteries are the primary constraint.
However, this is a textbook example of a vendor-driven narrative. The evidence provided consists entirely of internal specifications (800 GOPS, 1 mW/GOPS) and the promotion of a proprietary software ecosystem. There are no independent benchmarks, peer-reviewed comparisons against traditional ARM or RISC-V accelerators, or specific use-case data to validate the "efficiency" claim in real-world applications. The transition from a chip to a "reference design" (the AkidaTag) serves to move the conversation from raw silicon specs to a tangible product vision, effectively bypassing a critical discussion on the difficulty of training and deploying SNNs compared to standard backpropagation-based models.
The underlying paradigm is one of "Hardware-First AI," assuming that the primary bottleneck to AI ubiquity is energy efficiency, rather than algorithmic maturity or data quality. If this narrative holds, the benefit is a massive expansion of "invisible" AI in IoT devices. The cost is a potential increase in vendor lock-in, as developers must use specific conversion tools to move from open frameworks like PyTorch to proprietary hardware.
Patterns detected: ARC-0045 Authority Game
How would a coordinated influence campaign push this? A bad actor would flood technical forums with "leaked" benchmarks showing SNNs outperforming GPUs by orders of magnitude in specific niche tasks, creating a "fear of missing out" (FOMO) among engineers and forcing a pivot toward a specific proprietary architecture before the industry has standardized the software layer. The current content does not match this pattern; it is a standard promotional overview.
Bridge Questions:
1. How do the accuracy and latency of the converted TensorFlow/PyTorch models on SNN hardware compare to their original implementations on standard hardware?
2. What are the specific trade-offs in "effective GOPS" when moving from a simulation to a physical SNN implementation?
3. Beyond power efficiency, what fundamental capabilities does an SNN provide that a highly optimized traditional CNN or Transformer cannot?
Sentinel — Human
This text appears to be factual reporting on a technical product launch by providing specific details, which is characteristic of human-authored technical journalism.