ChipForge @ChipForge

As advanced LLMs accelerate chip design, the bottleneck is shifting from generating designs to verifying them. The challenge is no longer just how fast we can design, but how confidently we can validate.

Smarter AI needs smarter hardware. Cars that react in milliseconds. Robots that understand their surroundings. Smart devices making decisions in real time. All while keeping data private and power consumption low. That's where NPUs (Neural Processing Units) come in. Specialized chips built to run AI faster and more efficiently, right on the device, without constantly relying on the cloud. At ChipForge (SN108), we're pushing the boundaries of NPU design to power the next generation of Edge AI.

AI could push semiconductors toward a $2.3T market by 2030. The bottleneck isn’t just compute. It’s building better chips. That’s the opportunity for @ChipForge: engineers competing openly to design AI accelerators judged on real workloads. If it works, #SN108 on bittensor:native isn’t just riding the AI wave. It’s helping build the hardware behind it.

ChipForge @ChipForge

4 days ago

AI is everywhere, but it all runs on chips. McKinsey projects a $2.3T semiconductor market by 2030. Yet chip design remains costly and concentrated. ChipForge, Subnet 108 is bringing silicon design to Bittensor, opening competition to global talent to build practical chips.

2 5 30 2K 1

@pronounced_kyle We’re turning AI chip design into an open competition. Admittedly, the “personal fab” side needs work.

AI is moving into the devices around us. Cameras. Robots. Everyday devices that need small, efficient chips. ChipForge is opening AI chip design to a global network of builders on Bittensor SN108. Decentralized innovation. Intelligence at the edge. ↓

AI is everywhere, but it all runs on chips. McKinsey projects a $2.3T semiconductor market by 2030. Yet chip design remains costly and concentrated. ChipForge, Subnet 108 is bringing silicon design to Bittensor, opening competition to global talent to build practical chips. mckinsey.com/industries/sem…

@DaveSte82463893 We are going to start with Digital Chips

AI is about to be everywhere. Not just in your phone, but in everything around you. Cars that see and react in milliseconds. Robots that navigate the real world. Cameras that understand what they see. These devices can't always rely on the cloud. They need to make decisions right where they are. AI needs to run right where it's needed. Fast, private, and power-efficient. Making this possible requires a new generation of powerful, energy-efficient AI chips. At ChipForge (SN108), we're changing how those chips are designed.

@phoenix_ave8 Our focus is only on design. We can manufacture it through any foundary like tsmc. Most of the chip companies are fabless, means they only focus on design part, and manufacture it through tsmc

@Darkz4D Our focus is only on design of npu. We can manufacture it through any foundary like tsmc. Most of the chip companies are fabless, means they only focus on design part, and manufacture it through tsmc.

🚀 The ChipForge NPU Challenge is LIVE on SN108. Quest 001: Design an AI accelerator chip and the software that runs it. Your design will run four real AI models covering voice, vision and sensor tasks. Every answer must match the reference exactly. Then, it’s scored on chip area and speed. The challenge package includes the rules and evaluator, so you can test your design before submitting. Miners compete, validators verify, and the best designs earn rewards. This is our first step from open competition to real hardware. Get started: github.com/TatsuProject/C… Follow submissions, scores and the current winner: chipforge.io/leaderboard The challenge is open. Start building.

🚨 Chip design is one of the most closed industries on earth. Handful of giants hold the keys. A new subnet just put it on a public leaderboard. $TAO @ChipForge, SN108, from the @TatsuEcosystem team. It's formerly SN84 and now re-registered on mainnet. The first NPU challenge goes live today, Oct 6. Here is what they are doing: • Before: miners submitted Verilog for individual hardware blocks. ChipForge's own words: "good start, but limited scope" • They also ran RISC-V core design, checking performance, IPC, power and area • Now: miners design complete NPUs (neural processing units) and write the optimized kernels to run on them • Every design is tested on real ML models (MLPerf Tiny) and scored on speed, power and area • The hardware is open source. A good design doesn't die when the challenge ends Why I find this interesting: Bittensor works best when output can be scored objectively. A chip runs the model faster, cooler and smaller, or it doesn't. No opinions, just a number. The target is edge AI: phones, cameras, sensors and wearables. Tatsu's framing: everyone is fighting over datacenter GPUs, while the AI chips that ship in volume go into everyday devices. Benchmark score is not a taped-out chip. No silicon has been fabricated here, MLPerf Tiny models are small, scoring systems get gamed, and the account has 161 followers. This is early. Folks. Fair enough. But early is where a public leaderboard matters most. You can watch miners compete in the open. If those designs keep improving week over week, that's something the chip industry has never had: a public scoreboard. I'm watching the first challenge closely. Sources: ChipForge, Tatsu, chipforge.io, TatsuProject GitHub. $TAO / Not financial advice, always DYOR.

ChipForge @ChipForge

a week ago

The ChipForge NPU Challenge goes live tomorrow, October 6, at 9 AM EST AI is no longer limited to the data center. Intelligence is moving into everyday devices: the phones, cameras, sensors and wearables around us. This is edge AI, and it runs on a small, efficient chip called

2 6 34 6K 1

@phoenix_ave8 No issues, Its active now

Here's what makes ChipForge different from what we were before — and different from anything else on Bittensor. Previously (exSN84), miners submitted RTL (Verilog) code for individual hardware blocks. We verified it in our environment. Good start, but limited scope. Now, miners design complete NPU architectures and provide the optimized kernels needed to make the most of their hardware. ⚡ Speed — how fast does it run inference? 🔋 Power — how efficient is it? 📐 Area — how compact is the design? The best silicon wins. The network rewards it. This is how real chip companies compete. Now it's decentralized.

The ChipForge NPU Challenge goes live tomorrow, October 6, at 9 AM EST AI is no longer limited to the data center. Intelligence is moving into everyday devices: the phones, cameras, sensors and wearables around us. This is edge AI, and it runs on a small, efficient chip called an NPU (neural processing unit). Until now, designing one has taken large teams and large budgets. ChipForge opens this work to everyone. The challenge: design an AI accelerator chip and the software that runs it. How every design is judged: - It runs real AI models from MLPerf Tiny, the industry-standard benchmark for edge AI chips, covering voice, vision and sensor tasks. - Every result is checked bit for bit against the official reference. Only exact answers count. - Performance and chip area are measured with professional chip design tools. Miners compete to build the fastest and most efficient design. Validators run the same open evaluator, so every result can be reproduced. The best designs are rewarded, and each new winner has to clearly beat the last. The result is an open edge AI chip that gets better with every round. It's the first step on our path from open competition to real hardware. Live tomorrow. Start building.

The next breakthrough in AI hardware could come from anyone. Today, it never does. It stays locked inside a handful of giant companies. ChipForge changes that. A decentralized subnet on Bittensor where hardware designers compete to build the best AI accelerators. The better your design, the more reward you earn. Chip design, finally open to the world.

@taostacker @TatsuEcosystem @bittensor Thanks! We learned a lot from the last run and have made some pretty significant changes since then. This time we’re coming in with a much clearer plan and stronger validation. We’re looking forward to showing what we can do with SN108.