After decades of design innovation and process breakthroughs, MCUs have made great strides toward low cost, low power consumption, high processing efficiency, and high performance. With emerging application developments such as vehicle electrification, Industry 4.0, Edge AI, and industrial transformations, other great opportunities have arrived, and the pursual of higher-performance MCUs has become an industry trend.

Omdia, a technology research and advisory group, forecasted that the global MCU market revenue would reach $26.9 billion in 2022. Over the next five years, it will continue rising to a Compound Annual Growth Rate (CAGR) of 4.9%. The MCU market shipment forecast also shows that global MCU shipments will maintain a CAGR of 2.9% from 2022 to 2027.

▲ Global MCU Shipment Forecast

(Data Source: Omdia Market Tracker 2022)

Among the global MCU shipments, the 32-bit MCU market occupies an important percentage of those shipments, and its market share continues to rise. The 32-bit ultra-high performance MCUs contribute to the primary increment in this regard. Concurrently, communication, consumer electronics, industrial control, and automotive industries have integrated AI into their products. High computation rates and complex AI algorithms demand further semiconductor manufacturers to expand their ultra-high-performance product lines, thus accelerating the popularity of edge AI. As a leader in 32-bit general-purpose MCUs, GigaDevice is entering the ultra-high-performance market with the GD32H7 family (GD32H7).

Mr. Eric Jin, GigaDevice's Product Marketing Director, was interviewed to discuss the market advantages, target applications, and breadth of ecosystem development for the new GD32H7.

1. The GD32H7 is Riding High on Performance

The GD32H7 adopts a 600MHz Arm® Cortex®-M7 core based on Armv7E-M architecture. With a 6-stage pipeline architecture and support for high-bandwidth AXI and AHB bus interfaces, it provides higher main frequency and processing performance, achieving excellent 1552 DMIPS and 2888 CoreMark results. The GD32H7 delivers significantly higher performance to support advanced DSP, edge AI, and other high computing applications than other products.

Eric Jin explains that most Cortex-M4 core-based code can run directly on Cortex-M7, thus supporting seamless product migration. However, to take advantage of the core performance differences for full optimization, the software needs to be recompiled, and in many cases, some minor upgrades are required to take advantage of new features like Cache/TCM.

▲ GD32 Series Performance Comparison

The GD32H7 has up to 4MB of on-chip Flash and 1MB of SRAM to support large-capacity code storage. The unique TCM memory and L1 cache also significantly improve the efficiency of internal and external memory accesses. Among them, the 512KB ultra-large Tightly Coupled Memory (TCM), which can be freely configured as I-TCM or D-TCM, is used to store code and data that need to be accelerated to achieve zero-wait operation and improve system performance. It also integrates a 64KB L1-Cache (I-Cache, D-Cache), with storage speed close to the operating speed of CPU cores, solving the problem of excessive speed gap between CPU and memory and providing sufficient support for running complex operating systems and advanced algorithms.

The GD32H7 provides various secure encryption functions, including DES, Triple DES, AES, and hash algorithms. The integrated RTDEC module also secures data from external memory connected to the AXI or AHB bus, preventing threats during communication in the factory and on the field and ensuring the data security of IoT hardware.

The GD32H7 has significantly expanded peripheral resources and improved analog performance compared with existing high-performance products. The on-chip integration of two 14-bit ADCs with sampling rates up to 4MSPS and one 12-bit ADC with sampling rates up to 5.3MSPS provides high precision sampling rates and fast response in applications such as motor control and photovoltaic energy storage. Three channels of CAN-FD interfaces and two Ethernet controllers also offer excellent advantages for industrial network interface cards, inverters, and servers.

As per Eric Jin’s introduction, the GD32H7 will have three new series - GD32H737/757/759 to meet the demand for various applications.

GD32H737 series

· Support three channels of CAN 2.0B

· Available in BGA176 and LQFP176/144/100 packages

GD32H757 Series

· Support three channels of high-speed CAN-FD

· Available in BGA100 and LQFP144/100 packages

GD32H759 Series

· Support three channels of high-speed CAN-FD

· Available in BGA176 and LQFP176 multi-pin packages

▲ GD32H737/757/759 Series Selection

2. An Application Ecosystem that Enjoys the Dividends of Ultra-High Performance MCUs

The GD32H7 is well-prepared to be supported by GigaDevice’s ecosystem. The GD32 Eclipse IDE, GD32 All-In-One Programmer, and mainstream embedded development tools such as Arm KEIL, IAR, and SEGGER already support the GD32H7 products. In addition, for high-end application scenarios, GD32 is working with third parties to provide more algorithms and middleware resources, including signal processing, motor control, graphic display (GUI), voice recognition, and edge AI applications. Also, various development boards and learning kits are being launched with the new GD32H7 products, aiming to reduce application development complexity.

▲ GD32H759I-EVAL Full-function Development Board

Regarding the positioning difference with the existing products, Eric Jin said that the GD32H7 provides a development tool for more high-end needs and covers more target markets and application scenarios.

Human Machine Interface (HMI)

With the rapid iteration of display technology, various markets, such as household appliances, smart homes, and smart cities, are beginning to introduce intelligent HMIs. The emergence of diversified human-machine interaction applications such as voice, gesture, and touch accelerated the upgrade of MCUs in the new generation of IoT products.

The current pursuit of HMI display effects is getting more complex, which needs to be rendered into higher resolution and higher color depth, thus occupying more RAM space. With diversified interaction methods such as face recognition, gesture recognition, and voice control, customization demands are also emerging, and more GUIs, vector graphics, and visual effects are gradually added to the HMI interface, posing unprecedented challenges to the graphics performance requirements of the MCU.

The GD32H7 chip has a built-in TFT LCD driver, and with its 1MB SRAM capacity, it can handle the high memory requirements of higher image resolution in rendering and array transfer. Meanwhile, the GD32H7 integrates an Image Processing Accelerator (IPA), which supports 2D image overlay, rotation, scaling, and multiple color format conversions. It can meet the needs of advanced GUIs to achieve customizable intelligent HMIs.

Digital Energy

The integration model of "PV, storage, and charging" continues to be a hot topic in the digital energy field and is a typical scenario where digital information, artificial intelligence (AI), and photovoltaic (PV) technology are integrated. The MCU, the core component of the inverter, has become the key to maximizing the system’s efficiency. The PV inverter converts the variable DC voltage generated by the PV solar panels into AC power at utility frequency, which is fed back to the commercial transmission system. In this process, it is crucial to improve the accuracy of the sampling circuit to enhance the hardware efficiency further.

The GD32H7, as the master MCU can provide rich peripheral resources, including four 32-bit general-purpose timers, twelve 16-bit timers, four 64-bit/32-bit basic timers, two PWM advanced timers, and two 14-bit ADCs with sampling rates up to 4MSPS and one 12-bit ADC with sampling rates up to 5.3MSPS, to precisely generate PWM signals. It ensures the high accuracy and efficiency of the sampling circuit and meets the complex control requirements of the inverter.

To realize the bidirectional conversion and flow of electric energy, the energy storage inverter also needs to use complex algorithms to realize constant current and constant voltage control to ensure the grid’s safe and effective operation, which also puts high demands on the computing performance and hardware acceleration configuration of the MCU.

The GD32H7 provides a main frequency of up to 600MHz and a built-in independent DSP hardware accelerator and double-precision floating-point unit, as well as hardware algorithm accelerators such as TMU and FAC, which reduce the burden on the core while meeting the efficient operation of common FFT and FIR algorithms of inverters. For micro inverters, which have been growing significantly in recent years, and one-way grid-connected PV inverters, DSPs were previously mainly used to realize sinusoidal inverter control. Nowadays, with supply availability, cost reduction, and high-efficiency needs, many inverter manufacturers have started using high-performance MCUs instead of DSPs. The GD32H7 MCUs are an excellent choice.

Edge AI

As seen from the emergence of embedded intelligence applications in recent years, traditional general-purpose MCUs can hardly meet the growing arithmetic requirements of edge AI. MCUs must improve their main frequency, storage capacity, and hardware accelerators to integrate more lightweight AI algorithms into embedded applications. The GD32H7 MCUs provide just the right hardware resources to support these requirements.

In terms of performance, the 600MHz main frequency of the GD32H7 can ensure that the edge-side AI arithmetic requirements can be met without integrating additional hardware NPUs. Even when running at non-maximum main frequency, the accuracy of AI algorithms can still be guaranteed. Thanks to the advanced manufacturing process, the GD32H7 integrates more hardware resources, including hardware DSP, double-precision floating-point FPU, trigonometric TMU, and filtering algorithm FAC modules, which are crucial for AI loads such as image processing.

Regarding memory capacity, MobileNet or YOLO cannot run on MCUs due to the high resource and arithmetic requirements. Edge Impulse has optimized the FOMO machine learning object detection algorithm specifically for edge AI, with 1/30th of other algorithms' arithmetic and memory requirements. FOMO is highly scalable, supports running on Cortex-M7 core MCUs with an object detection efficiency of 30fps, and requires at least 100KB of cache space. FOMO can efficiently count a large number of small objects as long as the objects are similar in size and do not overlap. The GD32H7 is equipped with 4MB Flash and 1MB SRAM high-capacity memory to handle this type of load, further simplifying hardware costs and providing more advanced vision capabilities for embedded devices.

Industrial AI applications that can also benefit from the GD32H7 series are predictive maintenance solutions, where predictive models are integrated into algorithms based on sensor condition monitoring, anomaly detection, etc. For example, in the aforementioned digital energy application, intelligent arc pull detection based on AI technology can predict the generation of arcs in advance and effectively cut off grid-connected switches to eliminate arcs and prevent them before they occur. The intelligent introduction of machine learning (ML) enables PV inverters to continuously accumulate the characteristic parameters of detected arcs, thus optimizing and improving the ability to identify arcs and detect real arcs more accurately and reliably to ensure the safety of power transformation.

With the development of TinyML technology, more ML models will run on MCUs with small memory. More memory and arithmetic power mean higher accuracy and efficiency. The GD32H7 is equipped with 64KB L1 cache and 512KB TCM with large configurable zero-wait space, which can support lightweight AI algorithms to enhance the intelligence of industrial applications while balancing low-latency real-time processing and more accurate control.

3. The symbolic meaning of the GD32H7

The MCU market is a niche market, and manufacturers' pursual of high performance and low power consumption continues. The target applications for ultra-high-performance MCUs are still a blue ocean market. GigaDevice has seized the opportunity further and increased customers’ interest in pursuing high-end intelligent application development with the GD32H7.

The launch of the GD32H7 has far-reaching implications for the MCU market. As GigaDevice's first flagship MCU with an M7 core, it continues to strengthen the leading position of GigaDevice with its competitive price/performance advantage. The GD32H7 provides richer development options for high-end needs, supports innovative scenarios such as complex computing, AI, and multimedia technologies, and comprehensively promotes the application and popularity of MCUs in cutting-edge fields.

--Mr. Eric Jin, Product Marketing Director, GigaDevice

Eric Jin also said that GigaDevice is actively researching dual-core and multi-core processor architectures, and such ultra-high-performance MCU products will further increase the coverage of AI for edge devices in the future. In addition, future products will also integrate security technologies such as TrustZone and Lockstep to meet the security requirements of more advanced applications.

The ultra-high-performance GD32H7 MCUs will blur the traditional boundary between MCUs and MPUs and stimulate more market potential. The GD32H7 is part of market diversification and milestone products, signifying that the GD32 MCU market has officially entered the ultra-high-performance era.


Expand reading