The Zhitong Finance App learned that there are media reports that a Silicon Valley startup AI chip company fully supported by the US Department of Defense Advanced Research Projects Agency (DARPA) is seeking a financing scale of at least 70 million US dollars from the financial market to develop an ultra-efficient chip for artificial intelligence technology.

According to reports, this AI chip company called Encharge AI Inc. has joined the increasingly intense chip technology competition with the aim of manufacturing AI chips with better overall performance and providing the core driving force for the current wave of artificial intelligence technology. In the field of artificial intelligence, computing power has become a scarce and valuable resource, helping the market value of NVDA.US (NVDA.US), the AI chip leader, soar to more than $2 trillion, and prompted OpenAI CEO Sam Altman (Sam Altman) to seek billions of dollars from the financial market to meet OpenAI's huge demand for AI chips to increase the production capacity of AI chips needed by OpenAI around the world.

The US government, and the US military in particular, also seemed very interested. In addition to spending billions of dollars to stimulate the return of high-end manufacturing industries such as chips in the US, the US government is also spending money to support startups. EnCharge CEO Naveen Verma (Naveen Verma) has stated that enCharge “was born with the full support of Darpa and the entire Ministry of Defense.”

According to information, this startup AI chip company was derived from Velma's research project at Princeton University, and he is still a professor of electronics and computer engineering at Princeton University. “Even when we were divested, Darpa helped us find our first venture capitalists.”

EnCharge AI CEO Velma said the startup AI chip company is looking for new strategic investors for the next round of funding. However, he declined to comment further on the financing negotiations or the company's valuation. EnCharge's competitors include Netherlands-based Axelera AI BV and D-Matrix Corp., which have raised more than $160 million.

EnCharge AI supported by the US Department of Defense focuses on the field of “in-memory computing”

enCharge is one of many chip companies around the world dedicated to so-called “in-memory computing” (in-memory computing), which focuses on future chip structures. However, up to now, this has mainly been an academic research project, which can drastically reduce the energy consumed by chips during parallel computation at the artificial intelligence level and greatly improve computational efficiency. In addition to startups, leaders in chip manufacturing, such as TSMC, Intel, and Samsung Electronics, are also studying this technology. The purpose of this “in-memory calculation” chip is to save electricity and greatly increase the calculation phase rate by processing data at high speed in the place where the data is stored rather than transmitting the data to different locations.

Generally speaking, “in-memory computing” is a computing architecture aimed at improving computational efficiency and energy efficiency. It speeds up data processing by performing computational operations directly within the memory to reduce data transfer. This approach is particularly suitable for applications that require frequent access to large amounts of data, such as artificial intelligence (AI) training/inference and machine learning. Traditional computing architectures usually transfer data from the memory side to the processor for processing, and then send the results back to memory. Not only is this data movement process time-consuming, it also consumes a lot of electricity resources. In-memory computing attempts to break this bottleneck and reduce or even eliminate the need for data transmission by performing computational operations directly within the memory.

AI chips usually require a balance between high performance and low energy consumption. The “in-memory computing” model improves energy efficiency by reducing unnecessary data movement. In addition, many AI applications require extremely low latency, both on the end side and in the cloud. In-memory computing can provide faster response times to meet the needs of real-time processing.

Currently, the three mainstream potential paths for in-memory computing in the chip industry are SRAM, DRAM, and non-volatile memory (such as RRAM, MRAM, PCM, etc.). Among them, SRAM is the research method chosen by most chip companies. SRAM has the characteristics of high speed and low latency, and is suitable for caching and temporary data storage. In-memory calculation can use these characteristics of SRAM to achieve high-speed computing operations.

CEO Velma said that in addition to the growing demand for AI chips from private technology companies, the US government also has an urgent need for chips that can be processed quickly. For example, the ability to run military applications based on artificial intelligence technology in places with limited electricity (such as remote areas and airplanes) is an increasingly important defense requirement in the US government's view.

According to information, the first-generation chip developed by EnCharge was not specifically developed and designed for OpenAI, Anthropic, and other companies that train large models of artificial intelligence, but was specifically designed for applications using these large models to perform specialized high-precision predictions.

Velma said that the 50-person startup has already had customers evaluating its chips and technology and plans to start releasing new chip products manufactured by chip foundry giant TSM.US (TSM.US) next year.

According to reports, DARPA's recent funding for EnCharge has increased the total amount of funding provided by the US Department of Defense to about 23 million US dollars. Since EnCharge was quietly established in Santa Clara, California three years ago, the investment of the venture capital department under the US arms company Raytheon Technologies and other private investors is about double this amount.