News
As Artificial Intelligence (A.I.) and deep learning applications become more prevalent in an increasing number of sectors, the need for better performance, greater Deep Neural Network (D.N.N) model capacity, and lower power consumption is becoming increasingly important. But what does this have to do with analog computing? Let's answer that right now! Alongside A.I., D.N.N neural models are growing at an absurdly exponential rate. With these models, traditional digital processors struggle to deliver the required performance with low power consumption and adequate memory resources, especially for large models running on the edge.
Speak with one of our specialists and discover how Pix Force can transform your business
What is analog computing?
That is where analog computing comes in, allowing companies to achieve more performance with lower energy consumption in a small and cost-effective format. Analog computing has been researched for decades and offers two main benefits. First, it is incredibly efficient, as it leverages the memory element for neural network weight storage and computation, eliminating data movement. Second, analog computing has high performance, combined with low latency, making it suitable for calculating the hundreds of thousands of multiply-accumulate operations that occur in parallel during vector operations. Considering these two factors, analog computing is ideal for the latest edge-AI computing requirements.
Computational speed
The computational speeds and energy efficiency of analog compared to digital have long been promising. However, in addition to the incredible difficulty of developing this technology, one of the greatest historical impediments of analog computing has been its size, with analog chips and systems being very large and expensive, making the expansion of its use quite limited in the market. Today, the combination of flash memory and analog computing solves these challenges and you get a sum much greater than if you worked with the individual parts – this is analog in-memory computing based on the concept of In-Memory Computing (I.M.C), which brings energy efficiency and performance. This technology has arrived at a moment of memorable refinement and it now sets the stage for Artificial Intelligence computing in the coming decades.
The advantages of analog computing
The advantages of analog in-memory computing power (or analog computing, strictly speaking) at the lowest level come from the ability to perform massively parallel vector-matrix multiplications with parameters stored in flash memory arrays. Small electrical currents are directed through a flash memory array that stores reprogrammable neural network weights, and the result is captured using analog-to-digital converters (ADCs). By leveraging analog computing for the vast majority of inference operations, the analog-to-digital and digital-to-analog power overhead can be kept to a small fraction of the overall power budget, and a major reduction in computing power can thereby be achieved. There are also many second-order system-level effects that provide a major drop in power; for example, when the amount of data movement on the chip is several orders of magnitude lower, the system clock speed can be kept up to 10 times lower than competing systems, making the control processor's job much simpler. The use of analog computing processors for Edge-AI applications is a great fit for many different cases: drones equipped with high-definition cameras for computer vision applications that require running complex DNN neural models locally to provide immediate and relevant information to the control station. Processors using analog computing make it possible to deliver powerful AI processing that is also extremely power-efficient, so companies can deploy these networks on the drone, working in conjunction with a wide variety of computer vision applications. These applications include: • Monitoring agricultural yields,
• Inspection of critical infrastructure such as power lines,
• Cell towers, bridges, and wind farms,
• Fire damage inspection and coastal erosion analysis.
Applications
Another type of application for which analog computing will be ideal is in low-latency human pose estimation, which can be used in smart fitness devices, gaming, or even in industrial robotics.
Computing and Artificial Intelligence
Analog computing is the most ideal approach for AI processing due to its ability to operate using much less energy at a higher speed with faster frame rates. The extreme energy efficiency of analog computing technology will allow product designers to create incredibly powerful new features in small edge devices and will help reduce costs on a significant amount of wasted energy in enterprise AI applications. Harnessing the power of analog computing combined with flash memory, OEMs will be able to rethink what is possible with AI. Imagine the amount of exciting innovations we will see without the existing limitations on power, cost, and performance of edge AI applications.
Fabio Caraça
Fábio Caraça is the Chief Growth Officer at Pix Force. He leads Pix Force's transformation into a scalable SaaS operation, combining strategic vision, culture, and high-impact execution.
