| IP명 | Monolithic 3D-Integrated Neuromorphic PIM System with Self Rectifying Passive RRAM and CMOS Neurons for SNN | ||
|---|---|---|---|
| Category | Mixed | Application | 인공지능 (AI) |
| 실설계면적 | 4㎛ X 4㎛ | 공급 전압 | 1.0V |
| IP유형 | Hard IP | 동작속도 | 100MHz |
| 검증단계 | Silicon | 참여공정 | SS028-2502 |
| IP개요 | - TCAM Arrays: Perform parallel ternary (0, 1, don't-care) pattern matching between input spike patterns and stored class templates. Both SRAM-based and RRAM-based implementations enable comparative evaluation of associative memory performance. - MAC Arrays: Execute analog vector-matrix multiplication by accumulating currents from activated synapse cells. The dual implementation allows direct comparison of computational accuracy and energy efficiency between SRAM and RRAM technologies. - LIF Neuron Circuits: Analog CMOS circuits operating in subthreshold region. Each neuron integrates input current on a membrane capacitor, applies leakage decay, and generates output spikes when the threshold is exceeded. - Mesh NoC: A 2×2 mesh router structure providing scalable, address-based asynchronous communication between functional blocks with minimized latency and routing congestion. |
||
- 레이아웃 사진 -
|
|||