Modern AI applications tend to consume too much power. Because of this, they do not process data on the devices themselves, but send them to powerful cloud servers. If we learn how to run AI programs on fitness bands, drones, and other devices, we will have not only a broader application of technology but also an AI revolution. Another way of feeling this revolution is in the gaming and gambling spheres. They managed to bring future dreams into reality. You can try it on your own at PlayAmo login and try to win your jackpot. And it could become a reality thanks to a new NeuRRAM microchip from Stanford researchers.
Problems of Artificial Intelligence
Most modern AI systems perform processing "in the cloud". If you ask the image generator to draw a picture, the request is sent to the data center and the result is returned to the smartphone or computer via the Internet.
The process takes from 10 seconds to several minutes. It's quite normal. It is unlikely that the user needs to instantly get a picture. However, in other cases, this is more critical, for example, if AI monitors the state of the heart in real-time using data from a fitness bracelet. In this case, it is more efficient to perform processing directly on the device, rather than in the cloud. This approach is called edge computing.
The real-time feedback loop is only possible with something like edge computing. Edge computing has another benefit: it improves privacy. If medical information does not leave the wearable device, then attackers will not be able to intercept it during transmission.
So why do applications run in the cloud and not locally? The problem is that wireless devices have limited processing power and battery capacity. To run more advanced and energy-intensive AI programs, you have to use large cloud servers.
Microchip NeuRRAM
NeuRRAM is a microchip that will allow you to run advanced AI programs right on your devices. NeuRRAM eliminates the energy-intensive process of transferring data between the chip's compute module (where processing takes place) and the memory module (where data is stored).
The problem of data movement is analogous to an eight-hour trip to the office with a two-hour workday. NeuRRAM moves AI processing from a dedicated module to the memory itself, thanks to a new technology called Resistive Random Access Memory (RRAM).
This type of chip architecture is called Compute-in-Memory or CIM. It is used to improve the energy efficiency of microcircuits. However, this often comes at the expense of processing power, versatility, and accuracy.
The Stanford team overcame these limitations by changing the design, circuitry, and other features of the chip. NeuRRAM is twice as efficient as the best CIM devices, supports a variety of AI applications, and is as accurate as traditional chips in tasks such as character recognition and image classification.
Efficiency, versatility, and accuracy are important aspects of the wider adoption of the technology. However, it is not easy to implement them all at once. The key is to co-optimize the entire stack, from hardware to software.
Future Applications
Although NeuRAMM is just a proof of concept, the team plans to improve the device to handle more complex AI applications. If the chips were mass-produced, they would be ideal for wearable devices due to their low cost, small size, versatility, and energy efficiency, Wang said. Thanks to NeuRRAM, these devices will be able to handle AI applications that only cloud computing can handle today.
The technology will be useful not only in healthcare. The creators noted that NeuRRAM and other CIM chips have "virtually unlimited potential."
One day, these advanced microchips will give VR and AR headsets more power with less latency, and search and rescue drones will be able to analyze footage in real-time without quickly draining batteries.
These chips could also allow space rovers to autonomously explore other planets without having to wait minutes or hours to get directions from Earth.
Since information is useless if it takes minutes or seconds to be processed elsewhere, these chips could change our world into a more digital form. It's interesting how things have changed in the last decade. We wonder, how far it might lead us.
Special Content
