Blumind

Here’s why Blumind made my list of the Top 13 AI/ML Startups to Bet On Changing the World… The semiconductor industry is racing to deliver high-performance computing with low power consumption - essential for integrating AI into everyday personal devices. AI at the Edge (or inference) has the potential to usher in the “Age of AI Everywhere”, empowering billions worldwide and driving global human advancement. However, according to Niraj Mathur traditional digital architectures aren't the key to unlocking the full potential of inference. Before co-founding Blumind, he led product marketing at Rambus for high-speed connectivity and witnessed the skyrocketing demand for data center bandwidth. His customers explained that machine learning was driving the shift and Niraj realized ML would soon dictate all innovations in compute technology. He also recognized a critical flaw: modern processors were never built to handle ML efficiently. Blumind’s contrarian approach feels like a “better mousetrap”. In a recent interview, Niraj shared his insight, explaining, “We don’t think of what we do as compute, we fundamentally see neural networks as more of a signal processing problem. With computers, you have program memory, data memory, ALUs, a compiler that runs on these arithmetic units and coordinates the data movement… what the industry has done is try to shoehorn neural networks into this architecture, which leads to all the inefficiencies we see today." Niraj and co-founder John Gosson envision neural networks running far more efficiently on analog hardware designed to mimic the human brain. “The brain is a biochemical signal processor.” He said, “we’ve modelled our architecture around this approach rather than the digital Von-Neumann compute approach." If successful, Blumind could deliver performance improvements 100x-1000x better than the best processors on the market. With their first production chip now taped out, the solution couldn’t come at a better time. In the past six months, companies have hit a wall of efficiency and adding hardware only increases costs and power consumption – counter to the virtues of democratic, responsible AI. While I believe in Blumind, they face more than technical challenges. They'll also have to tackle the adoption challenge. As Niraj told me in our interview, “We have a hard time articulating how we’re different to people who are stuck in the old paradigm.” Digital architectures, though less efficient, are familiar. Potential customers have to understand WHY Bluemind’s approach is worth the risk. The same goes for talent acquisition. As Blumind scales, they need to convince engineers and developers that joining this venture is akin to being an early employee at NVIDIA with the potential for impact that will change the world. My advice to Niraj, Jon and Roger? Master your story until you can tell it in your sleep. There is power in persuasion. #semiconductorindustry #artificalintelligence #startups

On Tuesday, I launched my annual forecast of the Top 13 AI/ML Startups to Bet On Changing the World. Below are the seven companies that round out that list. Cerebras Systems - It’s easy to bet on a company when success seems like a foregone conclusion, however, no other AI startup I’m aware of has done so much to empower research institutions with access to accelerated computing that advances medical and life sciences than Cerebras. Andrew Feldman his team ARE changing the world and will continue to! Kneron - As a result of CEO 劉峻誠Albert Liu's relationship-building skills, Kenron has developed a breakthrough neural processing unit and are enabling the “Age of AI everywhere”. Albert has built a core team of amazing engineers and he’s systematically built partnerships with leading OEMs in automotive, consumer, and IoT markets. Lemurian Labs - A published mathematician and former Olympian, CEO Jay Dawani's career could have gone many directions but he dedicated himself to overcoming the challenges of developing in the Age of AI. Lemurian solved a 250 year-old math problem that created a breakthrough in computing efficiency, and developers worldwide will benefit. Blumind - While many AI companies are focused on digital architectures, Niraj Mathur and John Gosson started Blumind based on the thesis that analog computing is the key that will unlock the efficiency needed for AI at the edge. Fortune favors the bold and I believe this contrarian approach will lead them there. Rain AI - The Sam Altman-backed startup remained relatively obscure while they focused on developing their in-memory neuromorphic compute technology. However, they’ve recently closed a Series A and hired two chip development veterans from Meta and Apple to help, so I think this will be the year CEO William Passo brings the hard work of Rain’s co-founders to life. Etched - Co-founders Gavin Uberti and Chris Zhu were labelled “Harvard dropouts” when they launched in 2022 but if you listen to Uberti’s insights about AI, you realize that he deserves respect. Etched is making the right moves: raising capital and attracting key technical leaders to join their mission of challenging NVIDIA and making hardware for LLMs actually affordable. Taalas - Having worked with CEO Ljubisa Bajic during Tenstorrent’s formative years, I know his brilliance and believe he can usher in a paradigm shift in hardware built for AI. Taalas is inverting the development flow, using models to define hardware, instead of letting hardware dictate what models can do. This is a novel approach with great promise for increasing AI efficiency. Hailo AI - CEO Orr Danon and his cofounders gained deep insights into the importance of security and defense during their military service. Now, Hailo is building edge AI processors that power intelligent vision systems to accurately identify threats and keep people and buildings safer around the world. #semiconductorindustry #artificialintelligence #startups

I had the privilege to represent the Blumind team at the IEEE Circuits and Systems (ISCAS) 2024 conference with our debut paper on an extremely low power analog inference engine based on charge trap transistors and with no ADCs, for keyword spotting application. If that peaks your interest, please see our website at blumind.ai or reach out to us. The paper should be available soon too. PS. congrats to the organizers for a great conference! and what nice people are Singaporeans! might even surpass Canadians if I dare say :)