Various companies, from IBM to RAIN Neuromorphic, see the potential, but Mythic is first to market.
Alberto Romero, Analyst at Cambrian-AI, contributed to this article.
Mythic is an AI analog processor company conceived to overcome the increasing limitations of digital processors. Founded by Mike Henry and Dave Fick, and headquartered in Texas, Austin, and Redwood City, California, Mythic aims to solve the technical and physical bottlenecks that limit today’s processors by using analog computing in a world dominated by technology. digital technology. Mythic wants to show that, contrary to common belief, analog is not a relic of the past, but rather a promise for the future.
Two main problems inhibit the pace of digital hardware development: The end of Moore’s Law and the Von Neumann architecture. For 60 years we have enjoyed ever more powerful hardware, as predicted by Gordon Moore in 1965, but as we approach the theoretical minimum size of transistors, the well-used law of it seems to be coming to an end. Another well-known problem is the need in the Von Neumann architecture to move data from memory to processor and vice versa to perform calculations. This approach is increasingly being replaced by compute-in-memory (CIM) or near-memory computing approaches that significantly reduce memory bandwidth and latency while increasing performance.
The return of analog computing?
Mythic claims to have created a unique paradigm-changing solution that promises to address the limitations of digital while providing improved specifications compared to best-in-class digital solutions: an Analog Compute Engine (ACE). Historically, analog computers were superseded by digital ones due to the latter’s reduced cost and size, and their general-purpose nature. However, the current AI landscape is dominated by deep neural networks (DNNs) that do not require extreme precision and, more importantly, most of the computation goes into a single operation: matrix multiplication. The perfect opportunity for analog computing.
In addition, Mythic is exploiting the benefits of CIM and data flow architecture for impressive early results. They have taken CIM to the extreme by computing directly on flash memory cells. Their analog array processors take inputs as voltage, weights are stored as resistance, and output is the resulting current. Additionally, the data flow design keeps these processes running in parallel, enabling extremely fast and efficient computations while maintaining high performance. An intelligent combination of analog computing, CIM and data flow architecture defines Mythic ACE, the company’s main differentiating technology.
Legendary architecture leverages the low latency and low power consumption of analog computing
Mythic ACE Meets Edge AI Inference Requirements
Mythic’s technology promises high performance with very low power, ultra-low latency, low cost, and a small form factor. The basic element is its Analog Array Processor (AMP) which features an array of tiles, each containing the ACE supplemented with digital elements: SRAM, a vector SIMD drive, a NoC router, and a high-end RISC-V nanoprocessor. 32 bit. ACE’s innovative design eliminates the need for DDR DRAM, reducing latency, cost, and power consumption. AMP chips can be scaled, providing support for large or multiple models. Its first product, the single-chip M1076 AMP (76 AMP tiles) can handle many endpoint applications and can be scaled up to 4 AMPs or even 16 AMPs on a single PCI Express card, suitable for high performance edge servers . use.
The Mythic product family supports a wide range of edge deployments.
The hardware is complemented by a software stack that provides a seamless pipeline from graph (ONNX and PyTorch) to an AMP-ready package through optimization (including quantization to analog INT8) and compilation. Mythic’s platform also supports a library of out-of-the-box DNNs, including object detection/classification (YOLO, ResNet, etc.) and pose estimation models (OpenPose).
The company’s full-stack solution harnesses the potential of analog processors while maintaining features relevant to the digital world. Makes the M1076 AMP an excellent choice for handling AI workloads for inference at the edge faster and more efficiently (the company claims it provides “best-in-class TOPS/W”) than its all-digital counterparts . That, and the company’s broad offering of AI products and models, position it well to target fast-growing AI-focused markets such as video surveillance, smart home devices, AR/VR, drones and robotics.
So far, it appears that Mythic has turned an innovative idea into a promising technology to compete for edge inference AI. Now, let’s look at the numbers. The company claims that the M1076 AMP is powered by up to 25 TOPS running at around 3W. Compared to similar digital hardware, this is a reduction in power consumption of up to 10 times. And you can store up to 80 million pesos on the chip. The MP10304 Quad-AMP PCIe card can deliver up to 100 TOPS at 25W and store 320M pesos. When we compare these claims with those of many others, we can’t help but be impressed.
Conclusions
The success of analog AI will depend on achieving high density, high throughput, low latency, and high power efficiency, while delivering accurate predictions. Compared to pure digital implementations, analog circuits are inherently noisy, but despite this challenge, the benefits of analog computing become apparent as processors such as the M1076 can run larger DNN models that boast higher precision. , higher resolution or lower latency.
As Mythic continues to refine its hardware and software, we look forward to seeing benchmarks that can demonstrate the capabilities and power efficiency of the platform. But we’ve already seen enough to get excited about the potential of this unique approach.