We Don't Predict Latency.
We Schedule Through It.
Time-Based Scheduling is a fundamentally different architecture: it hides memory latency by knowing exactly when results will arrive, enabling sustained execution without speculation-driven stalls.
The scheduler knows. The pipeline never waits.
Conventional out-of-order processors speculate about when data will arrive and recover when they're wrong. TBS takes a different approach — at instruction issue time, the processor knows exactly when each result will be available, and schedules accordingly.
Designed to reduce dependence on branch misprediction recovery and reduce pipeline flushes — sustaining throughput where conventional designs stall.
Instruction timing is determined at issue — no runtime speculation or recovery needed.
Execution determinism by design.
| Execution determinism | Full / by design |
| Speculation overhead | Zero |
| Vector extension | RVV 1.0 |
| VLEN range | 128 – 2048 bits |
| Data path | 2× VLEN |
| Peak performance | 2.56 TOPS (5nm, single core) |
No wall between CPU and VPU.
The gap doesn't close. The processor stops caring.
At 10-cycle memory latency, SimplEx is 2.8× faster than best-in-class RISC-V CPU/VPU. At 100-cycle DRAM latency — real-world edge conditions — the advantage grows to 16×. SimplEx performance stays flat across the entire range.
No preload. No DMA. No dedicated SRAM.
Non-cacheable memory latency is the silent killer in embedded AI. SimplEx's predictive load mechanism delivers data on schedule — eliminating stalls before they occur, not recovering from them after.
Time-Based Microprocessors
In this white paper, Founder & CTO Dr. Thang Tran details a novel time-based architecture that achieves high-performance execution without the overhead of traditional dynamic schedulers.
7 pages · PDF
Get your free copy
See the architecture validated.
RTL-validated performance. Measured, not estimated.
View Performance Data