TrueNorth

This article covers the processor. For more meanings, see True North .

16 TrueNorth chips on a developer board

TrueNorth is a second generation neuromorphic processor from IBM , which was developed as part of the DARPA SyNAPSE program . The chip is inspired by the way the neocortex works and depicts it in a functionally simplified manner. It is not based on the Von Neumann architecture , but is still made from common materials and processes.

Technical specifications

Basic circuit diagram of a "Neurosynaptic Core"

True North is a The 4096 spike neural network Cores with a total of 1 million artificial neurons (the human brain has about 100 billion neurons ) and synapses 256 million and 400 Mebi SRAM with whom he synaptic operations per second per watt ( SOPS ) creates. It has a power density of and a power consumption of 70 mW to 100 mW. The chip works concurrently , non-clocked and fault-tolerant . TrueNorth-based systems are also highly scalable . ${\displaystyle 46\cdot 10^{9}}$ ${\displaystyle 20\,{\frac {\mathrm {mW} }{\mathrm {cm} ^{2}}}}$

The production takes place in a 28 nm process and the chips can be connected together as required. New technologies such as memristors are not used at True North.

application

TrueNorth is used in particular to simulate recurrent neural networks and is therefore ideal for applications in pattern analysis , such as B. image and speech recognition , as well as sensor and control technology . TrueNorth-based systems are to be integrated into Watson in the future.

In 2014, the Air Force Research Lab (AFRL) was the first to acquire a TrueNorth prototype. At the end of 2016 it was announced that a TrueNorth chip, compared to an NVIDIA Tegra K1 (on a Jetson TX-1 developer board), provided the same detection rate when detecting vehicles on the MSTAR data set, but only one-twentieth to one-thirtieth of the energy requirement would have. The high costs of TrueNorth - due to the lack of mass production - and the lack of suitable software libraries were seen as disadvantageous. However, the AFRL suggested that the TrueNorth chip would be recommended in environments with limited energy availability - such as in autonomous vehicles, drones and satellites.

Further development

The processor will be further developed into a cortical processor and will in future implement a hierarchical temporal memory (HTM) natively in hardware.

Web links

• Adam Clark Estes: IBM's New Brain-Like Chip Squeezes One Million Neurons Onto a Stamp. Gizmodo , August 7, 2014, accessed August 7, 2014 . 
• Alex Knapp: IBM Builds A Scalable Computer Chip Inspired By The Human Brain. In: Richard Dawkins Foundation . August 8, 2014, accessed August 10, 2014 . 
• Don Clark: Artificial brain: IBM develops new computer chip. The Wall Street Journal , August 11, 2014, accessed August 11, 2014 . 
• DOE CSGF 2015: SyNAPSE and Cortical Processor. In: YouTube. Krell Institute, August 20, 2015, accessed January 8, 2015 . 

swell

1. ↑ ^{a b c} Dharmendra S. Modha : Introducing a Brain-Inspired Computer: TrueNorth's neurons to revolutionize system architecture. IBM Research, accessed August 7, 2014 .  
2. ↑ Don Clark: brain chip with normal processor technology. Die Welt , August 11, 2014, accessed on August 12, 2014 . 
3. ^ Introducing a Brain-Inspired Computer. TrueNorth's neurons to revolutionize system architecture // IBM quote: "46 billion synaptic operations per second, per watt"
4. ↑ IBM introduces artificial brain. Retrieved October 1, 2014 .  
5. ↑ [1]
6. ^ Andrew Rosenblum: Air Force Tests IBM's Brain-Inspired Chip as an Aerial Tank Spotter. Technology Review , January 11, 2017, accessed January 13, 2017 .  
7. ↑ Werner Pluta: Cortical Processor: Darpa has brain computers developed. In: Golem. August 21, 2013, accessed January 8, 2015 .