Content-addressable memory (CAM) can compare a search word against many stored words in parallel, making it useful where fast lookup is important. But the parallel comparison circuitry and match-line operation can carry a substantial power cost. At the same time, memory nodes in scaled CMOS technologies can be vulnerable to single-event upsets (SEUs). A radiation-hardened CAM design that also avoids match-line precharge aims to address both concerns: reducing search energy while helping stored data recover after a modeled radiation event.
Why CAM design must balance search speed, energy, and reliability
A CAM array stores words in rows of cells. Each cell has a storage element—commonly SRAM—and comparison circuitry that checks its stored bit against the corresponding search input. The cells in a row contribute to a match line (ML): the line indicates whether the stored word matches the search word. Match-line sensing and subsequent encoding can then identify a matching row.
In a conventional NOR-type CAM, the match line is precharged before evaluation. A mismatch creates a path that discharges the line; a match leaves it charged. This approach supports fast parallel searching, but the precharge and evaluation behavior can consume power. In particular, the paper identifies a short-circuit current path during precharge under mismatch conditions as a source of additional power use.
Reliability adds another design constraint. An SEU occurs when charge from a radiation-induced event is sufficient to disturb a sensitive memory node. The resulting state change can corrupt the stored bit and, in turn, cause the CAM to report an incorrect match or miss. The study models this effect by injecting a double-exponential current pulse (DECP) at sensitive nodes and examining whether the memory state and match-line output remain correct.
What precharge-free CAM changes
Precharge-free architectures remove the match-line precharge phase, reducing the operations required for a search cycle. The paper discusses a self-controlled precharge-free (SCPF) CAM design, which uses stored data and comparison circuitry to control the match-line result without conventional precharging. Eliminating that phase can reduce search time and energy, although the SCPF design discussed in the study has its own power and delay trade-offs.
The proposed radiation-hardened precharge-free (RHPF) cell builds on this idea while changing both the storage circuit and parts of the comparison and access circuitry. It replaces the SCPF cell’s 6-transistor SRAM storage with a radiation-hardened memory (RHM) cell. Transmission gates provide access during writing, and a diode-connected transistor is included in the proposed circuit. The design also uses gated supply and ground elements to address leakage power.
How the RHPF cell handles writes and searches
During a write, word-line controls switch on the access transmission gates so complementary data can be written through the bit lines. For evaluation, those access gates are switched off, and complementary search inputs are compared with the stored data. The match or miss result depends on the charge developed at an internal node: the circuit drives the match line high for a match and low for a mismatch.
The transmission-gate comparison circuitry is intended to improve the voltage swing at that internal node, while the precharge-free operation avoids a separate match-line charging step. Together, these circuit choices are intended to improve energy use across the write and evaluation phases without giving up search performance. These are design objectives evaluated through circuit simulation, rather than claims of demonstrated performance in a manufactured device.
Why the storage cell matters for soft-error protection
The study compares NOR CAM cells using three types of SRAM storage: a conventional 6T cell, an NMOS-stack-based cell, and an RHM cell. The 6T version can flip its stored state after a modeled pulse at a sensitive node, with no recovery mechanism to restore the original value. The NMOS-stack version can recover from one direction of state change described in the paper, but not the other.
By contrast, the RHM cell is designed to recover from both 0-to-1 and 1-to-0 state changes caused by a single-node upset in the modeled conditions. When combined with the CAM comparison circuitry, this recovery helps preserve the correct match/miss output following a DECP strike. The distinction is important: protecting the storage node supports reliable lookup, not merely retention of a bit in isolation.
What the 45 nm simulations report
The proposed and comparison designs were simulated in Cadence Virtuoso using 45 nm CMOS technology. The reported comparison uses a 1 V supply. In the paper’s performance table, the proposed design has an average power of 0.1593 µW, a match-line delay of 5.66 ps, and energy of 0.212 fJ. The corresponding reported energy values are 0.92 fJ for the NOR CAM and 0.338 fJ for the SCPF CAM.
Based on those simulated energy values, the proposed design is reported to use 76.95% less energy than the NOR-type CAM and 37.27% less than the SCPF CAM. Its reported match-line delay is also lower than the comparison values in the table: 18.4 ps for NOR CAM and 22.7 ps for SCPF CAM. These figures describe the paper’s simulation setup and should be interpreted in that context, not as guaranteed results for other process nodes, layouts, or operating conditions.
The authors also report examining process corners, supply voltage, and temperature. In the corner simulations, the slow-slow condition produced the worst search delay, while the fast-fast condition produced the best. Such variation analysis is relevant when assessing whether a circuit’s behavior remains acceptable beyond a typical process condition.
Practical implications for memory and logic designers
For engineers evaluating CAM-based lookup, the work illustrates why search architecture and storage-cell resilience need to be considered together. Removing match-line precharge can reduce the number of steps in a search cycle, but a low-energy architecture must still produce a dependable match result. A hardened storage cell addresses a different part of the problem: preserving the data that the comparison logic relies on.
The reported RHPF results are simulation-based and specific to the evaluated design and 45 nm conditions. Applying the approach elsewhere would require validation against the target process, array organization, operating range, and implementation requirements. The practical takeaway is that precharge-free comparison and SEU-resilient storage can be combined at the cell level, offering a design direction for CAM systems where energy, search delay, and soft-error behavior all matter.