Discovered Materials retweeted
Discovered Materials' Akash Ramdas argues the next chip scaling frontier isn't smaller transistors, it's how many you can stack on top of each other: HBM is already 12 high DRAM going to 16 high
"We're starting to do this thing where we're taking a logic chip and then putting at's where materials can make a difference, because now you're actually thermally limited. The thing that's preventing your scaling is no longer how small is your transistor, but how many transistors can you put on top of each other?"
"The high bandwidth memory you'd have seen is a stack of dynamic random access memory. It's a 12-high dynamic random access memory stack right now. One of the roadmaps is to make this even higher, 16-high, and this is how you deal with memory bottlenecks in inference or training."
"The next stage of that is putting that HBM on top of a logic die. Both these technologies are going to be partially enabled by new materials."
"We're starting to do this thing where we're taking a logic chip and then putting at's where materials can make a difference, because now you're actually thermally limited. The thing that's preventing your scaling is no longer how small is your transistor, but how many transistors can you put on top of each other?"
"The high bandwidth memory you'd have seen is a stack of dynamic random access memory. It's a 12-high dynamic random access memory stack right now. One of the roadmaps is to make this even higher, 16-high, and this is how you deal with memory bottlenecks in inference or training."
"The next stage of that is putting that HBM on top of a logic die. Both these technologies are going to be partially enabled by new materials."
2
3
29
17,687
Discovered Materials retweeted
The most critical bottleneck in AI infrastructure right now isn't energy, chips, or software. It's heat, and no amount of better plumbing can reach the place where it forms.
Today, Lightspeed is partnering with Akash Ramdas (@Akash__Ramdas) and Advaith Sridhar founders of Discovered Materials (@discoveredmat), who are solving exactly this. We are leading their $9M seed round, joined by Y Combinator (@ycombinator) and angels including Paul Graham (@paulg), Gokul Rajaram (@gokulr), and Thariq Shihipar (@trq212).
The numbers behind the problem: Hopper ran 700W on one die. Blackwell runs ~1.2kW on two. Parts on the roadmap are headed past 4kW. And cooling doesn't scale linearly; double the heat and you need roughly 11x the cooling work. Beyond that, better pumps don't help. The constraint is the material itself.
Akash and Advaith have worked this problem from both ends: in Stanford's materials labs, deep in the atomic physics of heat, and building verifiers and harnesses at frontier AI startups.
They are building Discovered Materials to accelerate materials discovery and the lab-to-fab timeline. Its engine proposes 2,000+ candidate crystal structures a day, scores each for thermal performance in minutes, runs physics simulations on a shortlist, and synthesizes what survives, compressing quarters of materials R&D into days. They start with thermal interface materials and the destination is inside the fab itself, in a semiconductor market headed well past $1.5T.
Hemant Mohapatra (@MohapatraHemant) and Abhiram Tarimala (@AbhiramTarimala) lay out the full thesis, from the reticle limit to the cooling-power law to why the verifier is the moat, in our latest Substack post: lsip.substack.com/p/partnering-w…
Welcome to the Lightspeed family, Advaith and Akash.
Today, Lightspeed is partnering with Akash Ramdas (@Akash__Ramdas) and Advaith Sridhar founders of Discovered Materials (@discoveredmat), who are solving exactly this. We are leading their $9M seed round, joined by Y Combinator (@ycombinator) and angels including Paul Graham (@paulg), Gokul Rajaram (@gokulr), and Thariq Shihipar (@trq212).
The numbers behind the problem: Hopper ran 700W on one die. Blackwell runs ~1.2kW on two. Parts on the roadmap are headed past 4kW. And cooling doesn't scale linearly; double the heat and you need roughly 11x the cooling work. Beyond that, better pumps don't help. The constraint is the material itself.
Akash and Advaith have worked this problem from both ends: in Stanford's materials labs, deep in the atomic physics of heat, and building verifiers and harnesses at frontier AI startups.
They are building Discovered Materials to accelerate materials discovery and the lab-to-fab timeline. Its engine proposes 2,000+ candidate crystal structures a day, scores each for thermal performance in minutes, runs physics simulations on a shortlist, and synthesizes what survives, compressing quarters of materials R&D into days. They start with thermal interface materials and the destination is inside the fab itself, in a semiconductor market headed well past $1.5T.
Hemant Mohapatra (@MohapatraHemant) and Abhiram Tarimala (@AbhiramTarimala) lay out the full thesis, from the reticle limit to the cooling-power law to why the verifier is the moat, in our latest Substack post: lsip.substack.com/p/partnering-w…
Welcome to the Lightspeed family, Advaith and Akash.
0
5
37
4,369
Discovered Materials retweeted
Today, we're introducing @discoveredmat . We build AI scientists that discover new materials for semiconductor chips.
We’re starting by releasing our work - hundreds of new materials discovered using frontier AI models, along with our benchmark for tracking progress in this ery Bench.
To pursue our mission, we have raised $9 million from @LightspeedIndia (lead investor) , @ycombinator , @peakxvpartners , @paulg , @gokulr , @trq212 and many others.
🧵
We’re starting by releasing our work - hundreds of new materials discovered using frontier AI models, along with our benchmark for tracking progress in this ery Bench.
To pursue our mission, we have raised $9 million from @LightspeedIndia (lead investor) , @ycombinator , @peakxvpartners , @paulg , @gokulr , @trq212 and many others.
🧵
127
149
1,363
456,977