Renesas develops 3nm TCAM technology suitable for automotive SoCs
Renesas has announced a configurable ternary content-addressable memory (TCAM) built on a 3nm FinFET process. The new TCAM simultaneously delivers higher density, lower power and strengthened functional safety, making it suitable for automotive applications.
With the rapid expansion of 5G and cloud/edge computing, network traffic continues to surge, driving demand for large and diverse TCAM configurations such as 256-bit × 4,096 entries. Conventional scaling that relies solely on hard macros increases peripheral area due to more banks and repeaters and makes timing closure more difficult, while also raising search power. Automotive applications further require higher safety coverage to meet standards such as ISO 26262. Renesas addresses these challenges with the following innovations.
The newly developed TCAM hard macros are supported by a memory compiler in fine granularities—search key widths of 8–64 bits and entry depths of 32–128. Larger configurations (e.g., 256-bit × 4,096 entries) are realised by combining these hard macros with tool-driven soft-macro auto-generation to provide a configurable single macro that covers a wide range of use cases on one chip. This achieves industry-leading memory density of 5.27 Mb/mm².
Each hard macro integrates an all-mismatch detection circuit (Note 1) and executes a two-stage pipelined search. Based on the first-stage result, the second stage can be continued or halted to avoid unnecessary energy consumption. For example, in 64–256-bit × 512-entry configurations, the approach reduces search energy by:
Up to 71.1% with column-wise pipelined search (with key partitioning, >64-bit keys)
Up to 65.3% with row-wise pipelined search (without key partitioning, ≤64-bit keys)
In a 256-bit × 512-entry configuration, the design achieves low-power operation with search energy of 0.167 fJ/bit, and the distributed timing load enables a 1.7 GHz search clock. The resulting TCAM figure-of-merit (density × speed ÷ energy) reaches 53.8, exceeding prior work.
