IP Solution

본문 바로가기 주메뉴 바로가기

Solutions

IP Solution

SiliconGrid develops differentiated semiconductor IP to improve performance, power efficiency, reliability, and design productivity for advanced semiconductor applications.

IP Solution Key Visual

LPDDR-PIM solution

Low-Power, High-Performance LPDDR-PIM for Edge AI

SiliconGrid is developing LPDDR-PIM IP designed to address the memory bottleneck and power consumption caused by frequent data movement between processors and external memory.

Conventional computing architectures require continuous data transfers between CPU/GPU processing units and memory, increasing latency and energy consumption. While most PIM development has focused on HBM-based data-center applications, Edge AI requires a different approach optimized for strict power, thermal, and form-factor constraints.

SiliconGrid's LPDDR-PIM architecture brings computing closer to memory to reduce unnecessary data movement while improving AI inference efficiency and memory reliability.

LPDDR-PIM Visual

Key LPDDR-PIM Technologies

01

Optimized In-Memory Data Processing

02

Low-Power, High-Speed AI Computing

03

Enhanced Memory Reliability

Expected Benefits

  • Improved AI inference performance
  • Reduced data-movement overhead
  • Lower power consumption
  • Enhanced memory reliability
  • Optimized architecture for Edge AI applications

SiliconGrid is developing low-power, high-performance, and highly reliable LPDDR-PIM IP optimized for next-generation Edge AI systems

Analog IP performance optimization solution

Next-Generation IP Verification

Monte Carlo-Based Worst-Case Sample Selection & Re-Verification

At advanced process nodes, analog IP design must account for increasingly complex interactions among PVT variations, device random variation, layout dependency, and self-heating effects. Instead of repeatedly verifying every possible combination, vulnerable samples are selectively identified and re-verified while preserving the actual random variation, significantly reducing simulation cost.

Analog IP Verification Visual

Verification Challenges at Advanced Nodes

01

Advanced Process Scaling

Increasing sensitivity to process variation, device mismatch, layout-dependent effects, and self-heating.

02

Hundreds to Thousands of Verification Conditions

Rapidly increasing conditions due to combinations of PVT corners and repeated Monte Carlo simulations

03

Increasing Simulation Time & Cost

Limitations of the Conventional Approach : Repeated Monte Carlo simulations across all PVT combinations

SiliconGrid Verification Flow

  • Monte Carlo Analysis
  • Vulnerable Sample Selection
  • Device Modification & Random Variation Mapping
  • Precision Re-Verification

SiliconGrid selectively identifies vulnerable Monte Carlo samples and performs targeted re-verification while preserving device random variation, enabling accurate worst-case evaluation with significantly fewer simulation runs.

Verification Approach Comparison

Conventional Approach Visual

Conventional Approach

Full PVT Combinations × Repeated Monte Carlo Simulations

  • Large number of simulation runs
  • High computational cost
  • Long verification turnaround time
  • Difficulty maintaining consistent random-variation conditions after design modifications
SiliconGrid Approach Visual

SiliconGrid Approach

Selective Vulnerable-Sample Re-Verification

SiliconGrid selectively identifies vulnerable Monte Carlo samples and performs targeted PVT and post-layout re-verification while preserving device random variation. This approach enables accurate worst-case evaluation with significantly fewer simulation runs.

Target Simulation Workload: ~1/10 of Conventional Approaches*

*The actual reduction may vary depending on circuit characteristics, process technology, PVT conditions, and the number of Monte Carlo samples.