Analysis of NVIDIA's 2024 Patent Filing Trends

파인특허
September 30, 2024

NVIDIA is a global leader in technological innovation, driving advancements in fields such as artificial intelligence (AI), graphics processing units (GPUs), autonomous driving, and data centers. Through its cutting-edge GPU technology, the company plays a pivotal role across industries like gaming, AI, cloud computing, and autonomous vehicles. In this report, we will analyze NVIDIA’s patent filing trends to explore how the company leads in technological innovation. The patent data analysis was conducted using the keyword AP: (NVIDIA) in the Keywert search tool, with duplicate entries removed. This method allowed us to capture the global patent filing trends up to September 30, 2024.

1. NVIDIA's Patent Filing Trends by Year (up to 2024)

1995-1999:

The number of patent filings in the United States (US) stood out the most, especially with a sharp increase to 28 filings in 1999. Most patents were filed in the US, with few filings in other countries. Japan (JP) saw 2 filings in 1997 and 1 in 1999, with no filings in other nations during this period.

Early 2000s (2000-2005):

The US remained the dominant country for patent filings, with significant annual filings. There were some filings in Japan (JP), but far fewer than in the US. Korea (KR) saw its first filings during this period, with a gradual increase in numbers.

Early 2010s:

Patent filings in Korea (KR), Europe (EP), and China (CN) became more frequent. Notably, filings in Korea increased sharply, becoming a significant part of NVIDIA’s patent portfolio. While the US still had the most filings, other countries began to hold a more substantial share, indicating a growing global strategy.

Post-Mid-2010s:

Patent filings saw a dramatic rise, likely driven by NVIDIA’s expansion into new technology areas such as AI, autonomous driving, and high-performance computing, leading to a surge in related filings.

2. NVIDIA's Patent Filings by Country

The distribution of NVIDIA’s patent filings by country is as follows:

  • United States (US): 6,807 filings (largest share)
  • China (CN): 1,696 filings
  • International (WO): 552 filings
  • Korea (KR): 233 filings
  • Japan (JP): 212 filings
  • Europe (EP): 150 filings

The US holds the largest share of filings, followed by China and international filings.

3. Key IPC Categories of NVIDIA’s Patent Filings

  1. G09G-005/00 (243 filings): Display technology, particularly signal processing and control in display devices such as computer screens, monitors, and TVs. As a leader in GPU development, NVIDIA’s focus on ensuring GPU compatibility with high-performance display devices is evident.
  2. G06T-015/00 (235 filings): Digital image processing and analysis, including recognition and real-time video streaming. This is crucial for applications like autonomous driving and medical imaging.
  3. G06T-001/20 (223 filings): 3D graphics and computer graphics technology, crucial for industries like gaming, film, VR, and AR.
  4. G06N-003/08 (195 filings): Artificial neural networks and machine learning. NVIDIA’s GPUs are essential for neural network training and execution, with applications across autonomous vehicles, robotics, and natural language processing.
  5. H05K-007/20 (193 filings): Thermal management in electronic devices, crucial for maintaining GPU performance while managing heat.
  6. G06F-012/00 (178 filings): Data storage technologies, vital for optimizing large-scale data processing in data centers and cloud computing environments.
  7. G06T-015/06 (177 filings): Image reconstruction and enhancement, important in industries like gaming and autonomous vehicles.
  8. G06F-015/16 (139 filings): User interface (UI) technologies, improving user experience in high-resolution display environments.
  9. G06F-009/50 (130 filings): Task scheduling and management within computing systems, crucial for optimizing GPU parallel processing, especially in AI and deep learning models.
  10. G06N-003/04 (120 filings): Neural network learning methods, supporting faster neural network training.

NVIDIA’s patents reflect its leadership in graphics processing, AI, machine learning, data handling, and user interfaces. The company focuses heavily on GPU performance and advancements in AI and 3D graphics.

4. Major IPC Filings Over the Past Three Years

  1. G06T-015/06 (76 filings): Technologies improving image quality, often using AI-based image correction for industries like gaming, film, and autonomous driving.
  2. H05K-007/20 (72 filings): Thermal management techniques, critical for handling the heat generated by high-performance GPUs in environments like data centers.
  3. G06F-009/50 (69 filings): Task scheduling for efficient parallel processing in high-performance computing, essential in AI and deep learning.
  4. G06F-009/54 (68 filings): Control of data flow within computing systems, vital for enhancing performance in large-scale operations like data centers.
  5. G06N-003/08 (58 filings): AI and neural network learning technologies, used across a wide range of applications from autonomous driving to healthcare.

These areas underscore NVIDIA’s focus on enhancing AI, graphics, parallel processing, and efficient computing.

5. LDA Topic Classification

import pandas as pd
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.decomposition import LatentDirichletAllocation
from nltk.tokenize import RegexpTokenizer
from nltk.corpus import stopwords
import matplotlib.pyplot as plt
import numpy as np
import nltk


# NLTK 불용어 다운로드
nltk.download('stopwords')


# 특허 문헌에서 자주 사용되는 불용어 세트 정의
patent_stopwords = set([
    'method', 'apparatus', 'system', 'device', 'means', 'include', 'comprising', 'claim', 'wherein',
    'invention', 'embodiment', 'embodiments', 'described', 'present', 'application', 'may', 'can',
    'one', 'said', 'use', 'used', 'step', 'steps', 'provide', 'providing', 'example', 'plurality',
    'first', 'second', 'third', 'according', 'thereof', 'therein'
])


# 불용어 세트 (일반 영어 불용어 + 특허 관련 불용어)
stop_words = set(stopwords.words('english')).union(patent_stopwords)


# 텍스트 전처리 함수 정의
tokenizer = RegexpTokenizer(r'\w+')


def preprocess(text):
    tokens = tokenizer.tokenize(text.lower())  # 소문자로 변환 후 토큰화
    filtered_tokens = [word for word in tokens if word not in stop_words and len(word) > 2]  # 불용어 및 짧은 단어 제거
    return ' '.join(filtered_tokens)


# 데이터 로드
file_path = 'nvidia_en.xlsx'
data = pd.read_excel(file_path)


# 발명의 명칭과 요약을 결합한 텍스트 칼럼 생성
data['combined_text'] = data['발명의 명칭'].astype(str) + ' ' + data['요약'].astype(str)


# 텍스트 전처리
data['processed_text'] = data['combined_text'].apply(preprocess)


# 벡터화 (CountVectorizer 사용)
vectorizer = CountVectorizer(max_df=0.95, min_df=2, stop_words='english')
text_vectorized = vectorizer.fit_transform(data['processed_text'])


# 다양한 토픽 개수에 대해 LDA 모델을 학습하고 perplexity 값 계산
perplexities = []
topic_range = range(2, 21)  # 2~20개의 토픽 시도


for num_topics in topic_range:
    lda = LatentDirichletAllocation(n_components=num_topics, random_state=42)
    lda.fit(text_vectorized)
    perplexities.append(lda.perplexity(text_vectorized))  # perplexity 저장


# 토픽 개수에 따른 perplexity 그래프 그리기
plt.figure(figsize=(10, 5))
plt.plot(topic_range, perplexities, marker='o')
plt.xlabel("Number of Topics")
plt.ylabel("Perplexity")
plt.title("Perplexity by Number of Topics")
plt.show()


# 최적의 토픽 개수 선택
optimal_num_topics = topic_range[perplexities.index(min(perplexities))]
print(f"Optimal number of topics: {optimal_num_topics}")


# 최적의 토픽 수로 LDA 모델 학습
lda_final = LatentDirichletAllocation(n_components=optimal_num_topics, random_state=42)
document_topics = lda_final.fit_transform(text_vectorized)


# 상위 단어 출력 함수 정의
def display_topics(model, feature_names, no_top_words):
    for topic_idx, topic in enumerate(model.components_):
        print(f"Topic {topic_idx+1}:")
        print(" ".join([feature_names[i] for i in topic.argsort()[:-no_top_words - 1:-1]]))


# 최종 토픽 모델의 상위 10개 단어 출력
no_top_words = 10
tf_feature_names = vectorizer.get_feature_names_out()
display_topics(lda_final, tf_feature_names, no_top_words)


# 각 문서가 가장 많이 할당된 토픽을 계산 (가장 높은 확률의 토픽)
dominant_topic_per_doc = np.argmax(document_topics, axis=1)


# 토픽별 문서 개수 계산
topic_counts = np.bincount(dominant_topic_per_doc, minlength=optimal_num_topics)


print("\nTopic Counts (Number of documents assigned to each topic):")
for i, count in enumerate(topic_counts):
    print(f"Topic {i+1}: {count} documents")

Using the Latent Dirichlet Allocation (LDA) topic classification method, we identified several key topics across NVIDIA's patent filings. The dominant areas of innovation include.

  • Artificial Intelligence & Neural Networks (1,010 filings): NVIDIA’s dominant patent area, reflecting its extensive work in AI and machine learning.
  • GPU & Parallel Processing (850 filings): As the leader in GPU development, NVIDIA’s innovations here are essential for a range of applications from graphics to data processing.
  • Graphics & Texture Processing (768 filings): Reflecting NVIDIA’s continued leadership in enhancing graphics quality in industries like gaming and VR.
  • Memory Management & Virtual Memory (620 filings): Critical for optimizing performance in high-performance computing and data-intensive applications.
  • Graphics Programming & Hardware (500 filings): Related to programming interfaces and hardware advancements for GPU technology.

Other areas include various hardware and software technologies, reflecting NVIDIA’s wide-ranging innovations.

6. Conclusion

The analysis of NVIDIA's patent filings reveals that the company is continuously pushing the boundaries in cutting-edge technologies like GPU, AI, and parallel processing. NVIDIA plays a crucial role across industries, including AI, autonomous driving, and data centers, as reflected in its patent portfolio. Over the past three years, key areas of focus include AI, image processing, and parallel computing, supporting NVIDIA's strategy to expand its presence in the global market. This technological leadership is expected to keep NVIDIA at the forefront of innovation across multiple industries in the future.