Integration of eFPGA into microcontrollers is happening today now that this technology is available from multiple suppliers in 180nm to 16nm process nodes.
Microcontrollers today have an issue and an opportunity.
The issue is that at older process nodes, there are dozens of SKUs of microcontrollers with small variations in the type and number of serial I/Os and/or hardware accelerators. At 40nm, the mask costs start to rise substantially so doing dozens of variations is an expensive proposition.
The opportunity for microcontrollers is around the “glue” FPGA chips used by designers for decades. If these FPGA chips are integrated instead of being standalone, customers can significantly improve the cost, speed and power consumption of MCUs. This is a huge value proposition.
While not a new technology, embedded FPGA (eFPGA) is finally at the stage where it is ready to go mainstream with multiple suppliers, design wins in progress and proven silicon. Customers can leverage this technology in a number of ways, such as:
- A reconfigurable accelerator that can directly access on-chip buses, cache and I/O. One mask can cover multiple needs and customers can achieve higher performance.
- A reconfigurable I/O that can implement any serial I/O and can push low-level processing to the I/O block. This frees up the processor and improves responsiveness and battery life.
Microcontrollers often have dozens of variations to accommodate customer requirements for different combinations of serial I/Os: UART, USART, I2C, SPI and more.
With eFPGA, serial I/Os can now be programmed as needed. This enables MCU companies to save on mask costs and validation and provides customers with exactly the serial I/O they want, even variations on the standard versions.
There are multiple examples of programmable serial I/Os discussed in this app note. For example, in TSMC 40ULP, worst case frequency ranges from 20-50MHz depending on the type of serial I/O and clock; and the number of LUTs required (4-input LUTs in our T40ULP EFLX IP core) range from 72 to 590, depending on features required.
Initially, customers may not even realize they are using eFPGA because the manufacturer can program the eFPGA differently for each SKU.
This is just the start.
The next step is to use the eFPGA to process I/O so as to offload the MPU, improve performance and even lower power.
In another app note, you can see that using eFPGA to implement some simple, repetitive DSP functions can reduce power compared to implementing the same functions in the processor. The result is longer battery life.
Microcontrollers today sometimes have hardwired to offload the processors to improve performance. Examples of this are crypto-engines such as Advanced Encryption Standard (AES.)
Again, the microcontroller manufacturer can instead use eFPGA to implement multiple accelerators using one mask. This enables microcontrollers to be customized without adding incremental engineering time or cost. The customer may not even be aware of it. However, at some point, the manufacturer will likely want to enable the customer to program the embedded FPGA themselves, enabling any type of reconfigurable accelerator to be implemented.
Customers will then be able to reconfigure on the fly and change the accelerator function depending on the workload.
Microcontroller companies can also use embedded FPGA to implement various accelerator functions (such as AES, FFT, JPEG encode, SHA) with performance 30-130 times faster than an ARM processor. More details on this can be found in this app note.
Another option is to connect the eFPGA reconfigurable accelerator directly to GPIO: 8 bits, 16 bits, 32 bits or 64 bits. The manufacturer can also utilize this eFPGA to be an optional debugger as shown at this app note. This enables much more internal observation of microcontroller activity when the customer is trying to understand why their c-code/RTL combination is not getting the results they were expecting.
Integration of eFPGA into microcontrollers is happening today now that this technology is available from multiple suppliers in 180nm to 16nm process nodes. This will not only benefit manufacturers with lower engineering costs and faster time to market, but will also benefit microcontroller users with greater performance and flexibility in optimizing their systems.
—Tony Kozaczuk is the director of architecture solutions at Flex Logix.