Analog chips find new life in artificial intelligence


The need for speed is a hot topic among attendees at this week’s AI Hardware Summit – bigger AI language models, faster chips and more bandwidth for AI machines to make accurate predictions .

But some hardware startups are taking a throwback approach to AI computing to counter the more is better approach. Companies like Innatera, Rain Neuromorphics and others are creating silicon brains with analog circuitry to mimic brain functionality.

The brain is inherently analog, absorbing raw sensory data, and these chipmakers are trying to recreate how the brain’s neurons and synapses work in traditional analog circuitry.

Analog chips can be very good low-power sensing devices, especially for certain sound and visual applications, said Kevin Krewell, analyst at Tirias Research.

“Analog is a closer representation of how the brain works by using distributed memory cells to hold neuron weight or some other way to hold analog weight,” Krewell said.

AI and machine learning mostly rely on digital chips at the edge or in data centers. But there is a place for analog chips at the edge, like in smartphones or cars, which need instant intelligence without sending data to the cloud, which is used to deliver AI services.

“We are not aiming to replace the entire AI pipeline,” said Sumeet Kumar, CEO of Innatera Nanosystems BV, based in Rijkswijk, the Netherlands.

Innatera’s third-generation AI chip has 256 neurons and 65,000 synapses, which doesn’t seem like a lot compared to the human brain, which has 86 billion neurons and runs at around 20 watts. But Kumar said it was possible to create a fully connected recurring network and the chip could run on button batteries.

The chip is used by customers to run radar and audio applications, with performance competitive with other chips in the same class. The aim of the chip is to embed low levels of learning and inference on the device, which is seen as a big challenge for AI among show attendees.

“What we’re trying to do is recognize that as data moves from a sensor to the cloud, it’s actually transformed in multiple stages by different types of AI. And what we see very often , these are customers manipulating low-level sensor data in the cloud, which is completely unnecessary,” Kumar said.

The Innatera chip takes information from a sensor, which is converted into spikes, and the input content is encoded exactly when those spikes occur.

“That’s exactly how it goes in your brain. When you hear something, there’s… tiny hairs [cells] in your ear, which actually senses each frequency band and what the energy is in that band. And that hair [cells] will vibrate, produce spikes, which will then go to the rest of your auditory cortex. Essentially, we follow the exact same principle,” Kumar said.

Underlying this principle, inside brain neurons there are weak calcium ions and sodium ions, and these concentrations change over time. Innatera’s chip reproduces this same type of behavior using currents.

“We assess the amount of current entering the neuron and leaving the neuron. This is how we mimic the brain,” Kumar said.

The idea is not to disrupt the current flow of AI in the cloud, but to replace the current crop of edge AI chips that are not capable of making on-device decisions. The chip also reduces the process of converting analog signals to digital signals.

“You can’t really translate an analog signal over a long distance, because then you actually have degradation. We avoid that by converting that analog signal into a spike,” Kumar said.

The foundation of AI today is based on simulating the action of neurons in the brain using chips and digital techniques, which has met with great success. Based on advances in Moore’s Law, these digital circuits and networks have become larger and faster.

But analog has its problems. For example, it is more difficult to achieve consistency across analog chips with calibration issues like drift,

“Analog circuits and memory cells don’t scale like digital circuits. And most of the time, the analog eventually has to be converted to digital to interact with the rest of the system,” Krewell said.

Admittedly, the concept of neuromorphic chips is not new. Companies like Intel and IBM have developed brain-inspired chips, and universities are developing their own versions with analog circuitry. Intel and others have raised awareness of the difference between neuromorphic chips and conventional AI, but startups have felt the need to release their products as AI computational demands and power efficiency increase at an all-time high. unsustainable pace.

Another AI chip company, Rain Neuromorphics, said its brain-mimicking chip would be used in particle accelerators at Argonne National Laboratory.

In a presentation at the AI ​​Hardware Summit, the company didn’t provide many details about how the chip would be used, but company CEO Gordon Wilson said the chip would act as silicon brains that would help the research laboratory to study and draw conclusions about particle collisions.

The silicon brain will provide on-device intelligence to protect against sensor drift, which can result in erroneous data being sent to AI systems. The concept of sensor drift is similar to model drift in AI, in which bad data fed into a training model can send the AI ​​system off course.

Wilson claimed that the on-device chip capabilities are more power efficient than cloud AI.

“You need the ability to learn on the fly. You need the ability to train and adjust to that sensor drift to maintain the performance of that system,” Wilson said.

The first iteration of the Rain chip “essentially won’t be radically different from … other chip analogs or mixes,” Wilson said. But it will have the ability to learn, which will unlock more value.

Wilson pointed out different types of memory, like memristor circuits, providing the ability to learn. Memristors have been under development since the 1960s and pursued by HP (later to become HPE) for use in a megacomputer called The Machine, but the technology is still a novelty.

“Memristor serves as a memory resistor. It is a resistor that can adjust its resistance. It’s used like an artificial synapse,” Wilson said. In a brain, synapses are not required to be perfect, and the requirements will be different for Rain’s memristors.

Venture capitalist Sam Altman, known for his work in AI as CEO of OpenAI, invested $25 million in Rain Neuromorphics earlier this year.


About Author

Comments are closed.