Sagence AI Aims to Revolutionize AI Computing with Energy-Efficient Analog Chips

AI Sagence is working on something exciting: analog chips designed for AI applications. You might be wondering why this matters. Well, traditional graphics processing units, or GPUs, are energy hogs. They power most AI models but consume a lot of electricity. In fact, Goldman Sachs predicts a staggering 160% increase in electricity demand by 2030 due to the growing use of GPUs in data centers.

Vishal Sarin, an expert in analog and memory circuit design, believes this trend isn’t sustainable. After over a decade in the chip industry, he founded Sagence AI, previously known as Analog Inference. His goal? To create energy-efficient alternatives to GPUs.

Sarin explains, “The applications that could make practical AI computing truly pervasive are limited because the devices and systems processing the data can’t achieve the required performance. Our mission is to break those limitations in a responsible way.”

Sagence focuses on designing chips and systems for running AI models, along with the software needed to program these chips. While many companies are developing custom AI hardware, Sagence stands out by using analog chips instead of digital ones.

Most chips, including GPUs, store data digitally, using binary strings of ones and zeros. Analog chips, on the other hand, can represent data with a range of values. This flexibility is key.

Analog chips aren’t a new idea. They were popular from around 1935 to 1980, helping with projects like modeling the North American electrical grid. But now, as digital chips show their limitations, interest in analog technology is growing.

Digital chips often require hundreds of components to perform calculations that analog chips can handle with just a few modules. Plus, digital chips frequently need to shuttle data back and forth between memory and processors, creating bottlenecks.

As Sarin puts it, “All the leading legacy suppliers of AI silicon use this old architectural approach, and this is blocking the progress of AI adoption.”

Sagence’s analog chips are “in-memory” chips. They don’t transfer data from memory to processors, which means they can complete tasks more quickly. They also have a higher data density because they can store a variety of values.

However, analog technology does have its challenges. Achieving high precision can be tougher because it requires more accurate manufacturing. They can also be harder to program.

Sarin believes Sagence’s chips will complement digital chips, enhancing specialized applications in servers and mobile devices. He states, “Sagence products are designed to eliminate the power, cost, and latency issues inherent in GPU hardware while delivering high performance for AI applications.”

Sagence plans to bring its chips to market in 2025 and is already engaging with multiple customers. They aim to compete with other analog chip ventures like EnCharge and Mythic. Sarin adds, “We’re currently packaging our core technology into system-level products and ensuring that we fit into existing infrastructure and deployment scenarios.”

So far, Sagence has secured investments from notable backers, including Vinod Khosla and TDK Ventures, raising a total of $58 million since its founding six years ago. They’re also looking to raise more capital to expand their 75-person team.

Sarin explains, “Our cost structure is favorable because we’re not chasing performance goals by migrating to the newest manufacturing processes for our chips. That’s a big factor for us.”

The timing might be just right for Sagence. According to Crunchbase, funding for semiconductor startups is bouncing back after a slow 2023. From January to July, VC-backed chip startups raised nearly $5.3 billion, significantly more than the less than $8.8 billion raised in all of last year.

But let’s not forget, chipmaking is expensive. International sanctions and tariffs make things even more challenging. Plus, winning over customers who are already locked into ecosystems like Nvidia’s is no easy task. For instance, AI chipmaker Graphcore, which once raised nearly $700 million, filed for insolvency last year after struggling to establish a strong market presence.

To succeed, Sagence needs to prove its chips draw significantly less power and deliver higher efficiency than competitors. They also need to secure enough funding to manufacture at scale. The road ahead is challenging, but the potential is huge.