FlashX: Scaling Robotic Experience Reuse for Embodied Agents with CBA-Based NAND Flash
Organizations: Gwangju Institute of Science and Technology (GIST), Republic of Korea
Abstract
Embodied agents are emerging as a promising paradigm for complex robotic tasks, leveraging Vision-Language-Action (VLA) and World Action Models (WAMs) to generate actions through Diffusion Transformer (DiT)-based iterative denoising. However, existing models often struggle to generalize to out-of-distribution (OOD) environments. We observe that retrieving successful experiences and using them to warm-start denoising enables agents to better adapt to unfamiliar situations while reducing computational overhead. This creates a new opportunity to improve task accuracy and inference efficiency simultaneously, in contrast to conventional acceleration techniques that trade accuracy for computation. However, realizing this opportunity requires scaling experience banks, introducing substantial storage and search overhead that challenges the capacity, bandwidth, and power constraints of edge systems. We propose FlashX, a CMOS Directly Bonded to Array (CBA)-based NAND flash architecture for scalable experience reuse in embodied agents. FlashX exploits CBA's high internal bandwidth for parallel pSLC key search, employing interleaved Compute-in-Flash (CIF) engines to sustain sensing throughput while retaining dense TLC latent storage. Proprioception-aware clustering further reduces search overhead and enables incremental experience acquisition without global reclustering. Evaluations across VLA and WAM workloads demonstrate up to 12.24%p and 7.43%p improvements in task success under dense and compute-efficient configurations, respectively. FlashX achieves 2.57 higher search throughput than the strongest prior NAND search architecture under pSLC sensing, with a further 9.7 search reduction through clustering. This delivers 1.29--3.95 end-to-end speedup and 1.12--2.34 energy efficiency improvement over SSD-based retrieval.
Figures & tables
| Component | Parameter | AGX Orin | Orin NX |
|---|---|---|---|
| xPU | GPU | 1792 CUDA | 1024 CUDA |
| FP16 Dense (TFLOPS) | 27 | 13 | |
| INT8 2:4 (TOPS) | 108 | 50 | |
| Main Memory | Type | LPDDR5 | |
| Capacity (GB) | 32 | 16 | |
| Bandwidth (GB/s) | 204.8 (x256) | 102.4 (x128) | |
| Component | Implementation | Area | Storage |
| (mm 2 ) | (KB) | ||
| CIF Engine ( 12) | FP16 MAC, 200 MHz | 0.087 | – |
| Page Buffer | Reused LSB/CSB/MSB Latches | 0 | (192) |
| Query Buffer | SRAM | 0.013 | 4 |
| Search Ctrl. | Plane/Die Scheduling | 0.014 | – |
| Top-1 Reduction | Comparator Tree | 0.008 | – |
Memory-Augmented Agents
Memory-augmented agents use stored experience and knowledge to act coherently across long tasks, changing environments, and repeated interactions.
See all 369 papers